Coal mine disaster risk prediction method and system based on semantic recognition

Abstract

The present application provides a method and system for predicting coal mine disaster risks based on semantic recognition. The method includes the following steps: extracting coal mine feature data from historical coal mine accident data and coal mine non-accident data as a coal mine risk prediction training set based on a pre-built coal mine feature extraction model; inputting the coal mine risk prediction training set into a logistic regression model for training to obtain a coal mine risk prediction model; extracting coal mine hidden danger text information and coal mine feature vectors from measurement point data as coal mine risk assessment data based on the coal mine feature extraction model; calling the coal mine risk prediction model to perform risk prediction on the extracted coal mine risk assessment data. The present application fully considers the integration of coal mine text data and measurement point data, establishes a dynamic correlation between accident risk and hidden danger, tracks the risk of underground hidden dangers in coal mines in real time, improves the level of coal mine safety management and the efficiency of hidden danger disposal, blocks the key links of accidents, and reduces the risk of coal mine disaster accidents.

Description

Technical Field

[0001] The present application relates to the technical field of coal mine safety risk assessment, and in particular to a method and system for predicting coal mine disaster risks based on semantic recognition. Background Art

[0002] In recent years, as coal mines have strengthened their management in terms of standardization, the promotion and application of the coal mine "three-in-one" system (integrated management of coal mine risks, hidden dangers, and safety standardization) has changed the traditional hidden danger management mechanism and process. With the improvement of underground communication networks in coal mines and the use of explosion-proof mobile terminals, safety management personnel can check risk and hidden danger information on-site and enter it into the "three-in-one" system through explosion-proof mobile terminals. The system automatically transmits the hidden danger information to the safety management department, and each department conducts hidden danger rectification and supervision, realizing closed-loop management of hidden danger information. With the promotion and application of the "three-in-one" system in enterprises, large amounts of structured data and unstructured information are uploaded and aggregated, providing strong support for coal mines to deeply explore the correlations between related accident hazards through the application of natural language processing technology.

[0003] However, the existing coal mine disaster risk assessment and management methods have the following technical problems:

[0004] First, the analysis data is limited, primarily focusing on coal mine point data, without fully considering the integration of coal mine text data and point data. Analyzing coal mine text data is more complex than analyzing point data. While statistical analysis of point data yields most results, analyzing coal mine text data requires a combination of expert experience, manual annotation, and natural language processing methods and models.

[0005] Second, the correlation between coal mine hazards and accidents is currently low. While coal mines are currently investigating hidden dangers, there is no established correlation analysis model between various hidden dangers and accidents. This fails to better establish the logical relationship between accident risks and hidden dangers. Consequently, when investigating hidden dangers, it is impossible to dynamically and intuitively reflect the degree to which hidden dangers affect the corresponding accident risks.

[0006] Therefore, how to fully consider the integration of coal mine text data and measurement point data, establish a dynamic correlation between accident risks and hidden dangers, improve the level of coal mine safety management and the efficiency of hidden danger disposal, block the key links of accidents, and reduce the risk of coal mine disasters and accidents is a technical problem that needs to be solved urgently. Summary of the Invention

[0007] The purpose of this application is to provide a coal mine disaster risk prediction method and system based on semantic recognition, fully considering the integration of coal mine text data and measurement point data, establishing a dynamic correlation between accident risk and hidden dangers, improving the level of coal mine safety management and the efficiency of hidden danger disposal, blocking the key links of accidents, and reducing the risk of coal mine disaster accidents.

[0008] To achieve the above objectives, the present application provides a method for predicting coal mine disaster risks based on semantic recognition, which includes the following steps:

[0009] Obtain historical coal mine accident data and coal mine non-accident data;

[0010] According to the pre-built coal mine feature extraction model, coal mine feature data from historical coal mine accident data and coal mine non-accident data are extracted as a coal mine risk prediction training set;

[0011] Input the coal mine risk prediction training set into the logistic regression model for training to obtain the coal mine risk prediction model;

[0012] Obtain text data on coal mine hidden dangers and real-time data on measurement points;

[0013] According to the coal mine feature extraction model, the coal mine feature vectors in the coal mine hidden danger text data and the real-time data of the measurement points are extracted as the coal mine risk assessment data;

[0014] The coal mine risk prediction model is called to perform risk prediction on the extracted coal mine risk assessment data to obtain the coal mine risk prediction assessment results.

[0015] As described above, the method for pre-building a coal mine feature extraction model includes the following sub-steps:

[0016] Obtain hidden danger text vector training set;

[0017] Call the pre-set classification model to train the hidden danger text vector training set to obtain the text hidden danger vector corresponding to the hidden danger text;

[0018] Obtain the hidden danger vector of the measuring point, combine the hidden danger vector of the measuring point and the text hidden danger vector, and obtain the coal mine feature extraction vector.

[0019] As mentioned above, the coal mine feature extraction model extracts the coal mine feature vector, and the coal mine feature vector includes the measurement point hidden danger vector and the text hidden danger vector.

[0020] As described above, the method for obtaining a hidden danger text vector training set includes the following sub-steps:

[0021] Obtain a manually annotated hidden danger text training set;

[0022] Preprocess the hidden danger text training set;

[0023] The preprocessed hidden danger text training set is vectorized to obtain the hidden danger text vector training set.

[0024] As above, the method for obtaining the hidden danger vector of the measuring point is:

[0025] Obtain real-time data from coal mine measurement points;

[0026] The real-time data of the measuring point is judged, and the category corresponding to the risk hidden danger of the real-time data of the measuring point is used as the measuring point hidden danger vector.

[0027] As described above, the method for obtaining a coal mine risk prediction training set includes the following steps:

[0028] For each accident category, obtain the historical coal mine accident data corresponding to the accident category; and obtain the coal mine non-accident data;

[0029] According to the coal mine feature extraction model, the coal mine feature data from the historical coal mine accident data and coal mine non-accident data are extracted as the coal mine risk prediction training set;

[0030] Among them, accident categories include: gas accidents, water accidents, coal dust accidents and roof accidents.

[0031] As mentioned above, the coal mine accident information includes: coal mine hidden danger text data and measurement point hidden danger data. The coal mine hidden danger text data is used to extract text vectors, and the measurement point hidden danger data is used to extract measurement point vectors.

[0032] As above, a logistic regression model is used to learn the weights of the coal mine feature vectors and optimize the weights of the coal mine feature vectors.

[0033] As above, the coal mine risk assessment data is input into the coal mine risk prediction model; the coal mine risk prediction model outputs the predicted coal mine accident category.

[0034] This application also provides a coal mine disaster risk prediction system based on semantic recognition, which includes:

[0035] A training data acquisition module is used to obtain historical coal mine accident data and coal mine non-accident data; based on a pre-built coal mine feature extraction model, the coal mine feature data from the historical coal mine accident data and coal mine non-accident data is extracted as a coal mine risk prediction training set;

[0036] A model building module is used to input the coal mine risk prediction training set into the logistic regression model for training to obtain a coal mine risk prediction model;

[0037] Real-time data acquisition module, which obtains coal mine hidden danger text data and real-time data of measuring points;

[0038] The risk assessment data acquisition module is used to extract the coal mine feature vectors from the coal mine hidden danger text data and the real-time data of the measuring points according to the coal mine feature extraction model as the coal mine risk assessment data;

[0039] The risk prediction module is used to call the coal mine risk prediction model, perform risk prediction on the extracted coal mine risk assessment data, and obtain the coal mine risk prediction assessment results.

[0040] The beneficial effects achieved by this application are as follows:

[0041] (1) This application makes full use of the collected coal mine information data, including: hidden danger text data, measurement point real-time data, uses semantic recognition method to obtain coal mine text hidden danger vector, uses statistical method to obtain coal mine measurement point hidden danger vector, and fuses the two types of vectors to obtain coal mine feature vector, fully considering the fusion of coal mine text data and measurement point data, improving the accuracy of coal mine risk prediction, and reducing the probability of coal mine production safety accidents.

[0042] (2) This application establishes a correlation between hidden danger data and accident data. After the correlation is established, it can better analyze and mine the dynamic risk distribution of coal mine accidents and the correlation between hidden dangers, and provide data support for coal mines to reduce accident risks and deal with major risks in a timely manner. Specifically, this application makes full use of the collected accident data, which includes: gas accidents, water disasters, coal dust accidents and roof accidents, etc., and combines the coal mine feature vectors to use a logistic regression model to predict the coal mine accident risk in real time, and obtain prediction results, which are of great significance for coal mine risk prevention. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can also be obtained based on these drawings.

[0044] Figure 1 This is a flowchart of a coal mine disaster risk prediction method based on semantic recognition in an embodiment of the present application.

[0045] Figure 2 This is a flow chart of the method for pre-building a coal mine feature extraction model in an embodiment of the present application.

[0046] Figure 3 This is a flow chart of the method for obtaining a hidden danger text vector training set according to an embodiment of the present application.

[0047] Figure 4 Schematic diagram of the method for obtaining coal mine feature vectors according to an embodiment of the present application.

[0048] Figure 5 Schematic diagram of the method for coal mine risk prediction according to an embodiment of the present application.

[0049] Figure 6This is a structural diagram of a coal mine disaster risk prediction system based on semantic recognition in an embodiment of the present application.

[0050] Figure numerals: 10 - training data acquisition module; 20 - model building module; 30 - real-time data acquisition module; 50 - risk prediction module; 100 - coal mine disaster risk prediction system. DETAILED DESCRIPTION

[0051] The following is a clear and complete description of the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0052] Example 1

[0053] like Figure 1 As shown, the present application provides a coal mine disaster risk prediction method based on semantic recognition, which includes the following steps:

[0054] Step S1: pre-build a coal mine feature extraction model.

[0055] like Figure 2 As shown, step S1 includes the following sub-steps:

[0056] Step S110: Obtain a hidden danger text vector training set.

[0057] like Figure 3 As shown, step S110 includes the following sub-steps:

[0058] Step S111: Obtain a manually annotated hidden danger text training set.

[0059] Specifically, obtaining a coal mine hidden danger text training set includes: obtaining coal mine hidden danger text information in a recent period of time (for example, 60 days) and manually labeling the text information into categories.

[0060] Among them, the categories are those that are easy to cause coal mine disasters, which are divided by experts based on experience. The categories include: ignition source, insufficient air volume, abnormal mine water gushing, uncleaned water tank, water pump failure, pipeline failure, dust suppression failure, coal dust generation, untimely support, non-standard support work, failure of support work or reduced effect, and description of remaining hidden dangers, etc. A category label is set for each category, and the category label can be replaced by numbers, such as: 0, 1, 2, 3, etc.

[0061] Step S112: pre-processing the hidden danger text training set.

[0062] Specifically, the method for preprocessing the hidden danger text training set is: performing word segmentation and stop word removal on the hidden danger text training set.

[0063] Step S113: vectorize the preprocessed hidden danger text training set to obtain a hidden danger text vector training set.

[0064] Among them, the method for vectorizing the preprocessed hidden danger text training set is: using the TF-IDF (text vectorization) calculation method to calculate the text data in the hidden danger text training set to realize the vectorization processing of the text data in the hidden danger text training set.

[0065] Specifically, the TF-IDF calculation method is as follows:

[0066] TF-IDF=TF*IDF;

[0067]

[0068]

[0069] Among them, TF represents the term frequency, which represents the frequency of the term t appearing in the document; IDF represents the inverse document frequency index. The main idea of ​​IDF is that if there are fewer documents containing the term t, that is, the number of times the keyword appears in the hidden danger text is smaller, the larger the IDF is, which means that the term t has a good ability to distinguish categories, TF i,j Indicates the frequency of the i-th word in the j-th document; n i,j Indicates the number of times the i-th word appears in the j-th document; ∑ k n k,j Represents the total number of times all words appear in the jth document; idf i represents the inverse text frequency index of the i-th word; D represents the document; |D| represents the total number of documents; |{j,t i ∈D i} indicates that it contains the word t i The number of files.

[0070] Step S120 , calling a pre-set classification model to train the hidden danger text vector training set to obtain the text hidden danger vector corresponding to the hidden danger text.

[0071] Specifically, the hidden danger text vector training set is divided into a training set and a test set. The training set is used for training to obtain a text hidden danger classification model. The test set is used to test the text hidden danger classification model, optimize the text hidden danger vector, and obtain the optimized text hidden danger vector. The text hidden danger vector includes hidden danger features of multiple categories.

[0072] Preferably, the feature vectors and category labels (Label) in the hidden danger text vector training set are divided into a training set and a test set in a ratio of 8:2.

[0073] The training method of the text hidden danger classification model is: use SVM binary classifier for classification and train the text hidden danger classification model. When encountering multi-classification problems, a one-to-many strategy can be adopted to classify each category into one category and other categories into another category. A total of U SVM binary classifiers need to be constructed.

[0074] Using the kernel function, the optimization problem of solving the optimal classification hyperplane is transformed into the following minimization problem.

[0075] The minimization problem is:

[0076]

[0077]

[0078] Where K represents the kernel function; X represents the training features; C represents the penalty factor to prevent overfitting; l represents the total number of training features; n represents the number of times the keyword appears in the hidden danger text; y represents the category label; T represents transposition; st represents the condition that is met; p, q and k are parameters; α p and α q Indicates the parameters selected in the range of 0-C; ​​y p represents the p-th category label; y q represents the qth category label; X p represents the pth training feature; X q represents the qth training feature; α=[α1,α2,......,α l ].

[0079] The kernel function is used to transform low-dimensional space into high-dimensional space. Common kernel functions include RBF kernel, linear kernel, polynomial kernel, and Sigmoid kernel.

[0080] Step S130 , obtaining a hidden danger vector of a measuring point, combining the hidden danger vector of the measuring point with the hidden danger vector of the text, and obtaining a coal mine feature extraction model.

[0081] Specifically, the method for obtaining the hidden danger vector of the measuring point is:

[0082] Real-time data from coal mine measurement points is obtained. This data includes information about gas, oxygen, coal dust, wind speed, and mine pressure. Statistical analysis is performed on each measurement point to determine whether the data corresponds to a high-risk risk. If the data corresponds to a high-risk risk, meaning it is a potential risk, the corresponding category is used as the measurement point's potential risk vector. Otherwise, the corresponding category is not used as the measurement point's potential risk vector.

[0083] Specifically, the hidden danger vector of the measuring point and the hidden danger vector of the text are combined to obtain a coal mine feature extraction model, which is used to extract the coal mine feature vector.

[0084] The coal mine feature extraction model extracts coal mine feature vectors, that is, extracts measurement point hidden danger vectors and text hidden danger vectors. Specifically, if the extracted data contains data that is consistent with the measurement point hidden danger vector or text hidden danger vector in the coal mine feature extraction model, the consistent data is extracted and used as the coal mine feature vector.

[0085] like Figure 4 The figure shows a flow chart of a method for obtaining a coal mine feature vector. The method for obtaining a coal mine feature vector comprises the following steps: obtaining a text hidden danger vector and a measurement point hidden danger vector, and combining the text hidden danger vector and the measurement point hidden danger vector into a coal mine feature vector.

[0086] like Figure 4 As shown, the method for obtaining text hidden danger vectors is as follows: obtain a manually annotated hidden danger text training set, perform word segmentation and stop word removal on the obtained hidden danger text training set to obtain a preprocessed hidden danger text training set, perform vectorization on the hidden danger text training set to obtain a hidden danger text vector training set; train an SVM classification model based on the hidden danger text vector training set, and obtain a file hidden danger vector based on the hidden danger text test set and the trained SVM classification model.

[0087] The method for obtaining hidden danger vectors of measuring points is to perform statistical analysis on the real-time data of measuring points (gas, oxygen, coal dust, wind speed, mine pressure, etc.) to obtain hidden danger vectors of measuring points that are prone to high risks.

[0088] Step S2: Obtain historical coal mine accident data and coal mine non-accident data, and extract coal mine feature data from the historical coal mine accident data and coal mine non-accident data according to the coal mine feature extraction model as a coal mine risk prediction training set.

[0089] Specifically, for each accident category, the historical coal mine accident information corresponding to the accident category is obtained, and the coal mine non-accident data is obtained. According to the coal mine feature extraction model, the coal mine feature data in the historical coal mine accident data and the coal mine non-accident data are extracted as the coal mine risk prediction training set.

[0090] Among them, accident categories include: gas accidents, water accidents, coal dust accidents and roof accidents.

[0091] Step S2 includes the following sub-steps:

[0092] Step S210: For each accident category, obtain coal mine accident information within a recent period (eg, the past six months).

[0093] Coal mine accident information includes: coal mine hidden danger text data and measurement point hidden danger data. The coal mine hidden danger text data is used to extract text hidden danger categories, and the measurement point hidden danger data is used to extract measurement point hidden danger categories.

[0094] Step S220: extracting coal mine non-accident data in proportion from historical coal mine data.

[0095] Negative examples of non-accident data were randomly sampled at a ratio of 1:5, where positive examples represent accident data and negative examples represent non-accident data. In other words, the non-accident data was sampled at a ratio of 1:5.

[0096] Step S230: extracting coal mine feature data from the coal mine accident data and the coal mine non-accident data through a coal mine feature extraction module, and obtaining a coal mine risk prediction training set based on the coal mine feature data.

[0097] Coal mine characteristic data includes textual hazard classifications and measurement point hazard classifications. The extracted coal mine accident data and coal mine non-accident data are divided into a coal mine risk prediction training set and a coal mine risk prediction test set. The coal mine risk prediction training set is used to train a coal mine risk prediction model, while the coal mine risk prediction test set is used to test and optimize the coal mine risk prediction model.

[0098] Step S3: input the coal mine risk prediction training set into the logistic regression model for training to obtain a coal mine risk prediction model.

[0099] Among them, the logistic regression basic model is a linear regression model normalized by the Sigmoid function (logistic equation), which is a machine learning method used to solve binary classification problems.

[0100] The assumed functional form of the logistic regression model is as follows:

[0101]

[0102] Among them, h θ (x) represents the hypothetical function; x is the input, θ is the required parameter; e = 2.718.

[0103] The assumptions made by the logistic regression model are as follows:

[0104]

[0105] Wherein, g() represents a logic function; P(y=1|x; θ) represents the possibility of calculating the output variable y equal to 1 based on the selected parameter θ; and T represents transposition.

[0106] The decision function corresponding to the logistic regression model is:

[0107] y * =1,ifP(y=1|x)>0.5;

[0108] Among them, y * Represents a decision function; if represents if; x is the input; y is the output variable.

[0109] In the logistic regression model, the most commonly used cost function is cross entropy; in the process of optimizing parameters, the most commonly used method is gradient descent.

[0110] The logistic regression model is used to learn the weights of coal mine feature vectors, optimize the weights of coal mine feature vectors, and obtain a coal mine risk prediction model to achieve better prediction results.

[0111] Specifically, the weight of the coal mine feature vector is used as the parameter θ, and the coal mine feature vector is used as input. The parameter θ that satisfies the decision function is calculated as the weight of the optimized coal mine feature vector. The weight of the optimized coal mine feature vector is used as the weight parameter in the coal mine risk prediction model to better predict coal mine risks.

[0112] Step S4: Obtain coal mine hidden danger text data and measurement point real-time data.

[0113] Step S5: extracting coal mine feature vectors from the coal mine hidden danger text data and the measurement point real-time data according to the coal mine feature extraction model as coal mine risk assessment data.

[0114] In other words, the measuring point hidden danger vector and text hidden danger vector are extracted as coal mine risk assessment data.

[0115] Step S6: calling the coal mine risk prediction model, performing risk prediction on the extracted coal mine risk assessment data, and obtaining the coal mine risk prediction result.

[0116] Step S6 includes the following sub-steps:

[0117] Step S610: input the coal mine risk assessment data into the coal mine risk prediction model.

[0118] In step S620, the coal mine risk prediction model outputs the predicted coal mine accident category, so that corresponding preventive measures can be taken according to the predicted coal mine accident category.

[0119] like Figure 5 As shown in FIG, the method for obtaining the coal mine risk prediction results is: obtaining the coal mine risk prediction evaluation results based on the trained LR logistic regression model and the coal mine test set.

[0120] like Figure 5 As shown in the figure, the training method of the LR logistic regression model is: obtain coal mine accident data and coal mine non-accident data, extract the coal mine feature vectors therein as the coal mine risk prediction training set, and obtain the LR logistic regression model based on the coal mine risk prediction training set. The LR logistic regression model is used to predict coal mine risks.

[0121] Example 2

[0122] like Figure 6 As shown, the present application provides a coal mine disaster risk prediction system 100 based on semantic recognition, which includes:

[0123] The training data acquisition module 10 is used to obtain historical coal mine accident data and coal mine non-accident data; according to the pre-built coal mine feature extraction model, the coal mine feature data in the historical coal mine accident data and coal mine non-accident data are extracted as a coal mine risk prediction training set;

[0124] The model building module 20 is used to input the coal mine risk prediction training set into the logistic regression model for training to obtain a coal mine risk prediction model;

[0125] Real-time data acquisition module 30, acquires coal mine hidden danger text data and measurement point real-time data;

[0126] The risk assessment data acquisition module 40 is used to extract the coal mine feature vectors from the coal mine hidden danger text data and the measurement point real-time data according to the coal mine feature extraction model as the coal mine risk assessment data;

[0127] The risk prediction module 50 is used to call the coal mine risk prediction model, perform risk prediction on the extracted coal mine risk assessment data, and obtain the coal mine risk prediction assessment result.

[0128] The beneficial effects achieved by this application are as follows:

[0129] (1) This application makes full use of the collected coal mine information data, including: hidden danger text data, measurement point real-time data, uses semantic recognition method to obtain coal mine text hidden danger vector, uses statistical method to obtain coal mine measurement point hidden danger vector, and fuses the two types of vectors to obtain coal mine feature vector, fully considering the fusion of coal mine text data and measurement point data, improving the accuracy of coal mine risk prediction, and reducing the probability of coal mine production safety accidents.

[0130] (2) This application establishes a correlation between hidden danger data and accident data. After the correlation is established, it can better analyze and mine the dynamic risk distribution of coal mine accidents and the correlation between hidden dangers, and provide data support for coal mines to reduce accident risks and deal with major risks in a timely manner. Specifically, this application makes full use of the collected accident data, which includes: gas accidents, water disasters, coal dust accidents and roof accidents, etc., and combines the coal mine feature vectors to use a logistic regression model to predict the coal mine accident risk in real time, and obtain prediction results, which are of great significance for coal mine risk prevention.

[0131] The foregoing is merely an embodiment of the present invention and is not intended to limit the present invention. It will be apparent to those skilled in the art that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are intended to be included within the scope of the claims of the present invention.

Claims

1. A coal mine disaster risk prediction method based on semantic recognition, characterized in that: The method comprises the following steps: Obtain historical coal mine accident data and coal mine non-accident data; According to the pre-built coal mine feature extraction model, coal mine feature data from historical coal mine accident data and coal mine non-accident data are extracted as a coal mine risk prediction training set; Input the coal mine risk prediction training set into the logistic regression model for training to obtain the coal mine risk prediction model; Obtain text data on coal mine hidden dangers and real-time data on measurement points; According to the coal mine feature extraction model, the coal mine feature vectors in the coal mine hidden danger text data and the real-time data of the measurement points are extracted as the coal mine risk assessment data; Calling the coal mine risk prediction model, performing risk prediction on the extracted coal mine risk assessment data, and obtaining the coal mine risk prediction assessment results; The method for pre-building a coal mine feature extraction model includes the following sub-steps: Obtain hidden danger text vector training set; Divide the hidden danger text vector training set into a training set and a test set, use the training set to train the SVM classification model to obtain the text hidden danger classification model, use the test set to test the text hidden danger classification model, optimize the text hidden danger vector, and obtain the optimized text hidden danger vector; Obtaining the hidden danger vector of the measuring point, combining the hidden danger vector of the measuring point with the hidden danger vector of the text, and obtaining the coal mine feature vector; The method for obtaining a hidden danger text vector training set includes the following sub-steps: Obtain a manually annotated hidden danger text training set; Preprocess the hidden danger text training set; Performing vectorization processing on the preprocessed hidden danger text training set to obtain a hidden danger text vector training set; The method for vectorizing the pre-processed hidden danger text training set is as follows: using the TF-IDF calculation method to calculate the text data in the hidden danger text training set, thereby realizing the vectorization of the text data in the hidden danger text training set; TF-IDF is calculated as follows: TF-IDF=TF*IDF; Among them, TF represents the term frequency, which represents the frequency of the term t appearing in the document; IDF represents the inverse document frequency index. The main idea of ​​IDF is that if there are fewer documents containing the term t, that is, the number of times the keyword appears in the hidden danger text is smaller, the larger the IDF is, which means that the term t has a good ability to distinguish categories, TF i,j Indicates the frequency of the i-th word in the j-th document; n i,j Indicates the number of times the i-th word appears in the j-th document; ∑ k n k,j Represents the total number of times all words appear in the jth document; idf i represents the inverse text frequency index of the i-th word; D represents the document; |D| represents the total number of documents; |{j,t i ∈D i }| indicates that it contains the word t i the number of files; Among them, the method for obtaining the hidden danger vector of the measuring point is: Obtain real-time data from coal mine measurement points, including gas, oxygen, coal dust, wind speed, and mine pressure. Perform statistical judgment on the real-time data of the measuring points, and take the category corresponding to the risk hidden danger of the real-time data of the measuring points as the measuring point hidden danger vector.

2. The method for predicting coal mine disaster risk based on semantic recognition according to claim 1, characterized in that: The method for obtaining a coal mine risk prediction training set includes the following steps: For each accident category, obtain the historical coal mine accident data corresponding to the accident category; and obtain the coal mine non-accident data; According to the coal mine feature extraction model, the coal mine feature data from the historical coal mine accident data and coal mine non-accident data are extracted as the coal mine risk prediction training set; Among them, accident categories include: gas accidents, water accidents, coal dust accidents and roof accidents.

3. The method for predicting coal mine disaster risk based on semantic recognition according to claim 1, characterized in that: Coal mine accident data includes: coal mine text hidden danger data and measurement point hidden danger data. The coal mine text hidden danger data is used to extract text hidden danger vectors, and the measurement point hidden danger data is used to extract measurement point hidden danger vectors.

4. The method for predicting coal mine disaster risk based on semantic recognition according to claim 1, characterized in that: The logistic regression model is used to learn the weights of coal mine feature vectors and optimize the weights of coal mine feature vectors.

5. The method for predicting coal mine disaster risk based on semantic recognition according to claim 1, characterized in that: The coal mine risk assessment data is input into the coal mine risk prediction model; the coal mine risk prediction model outputs the predicted coal mine accident category.

6. A coal mine disaster risk prediction system based on semantic recognition, characterized in that: The system includes: A training data acquisition module is used to obtain historical coal mine accident data and coal mine non-accident data; based on a pre-built coal mine feature extraction model, the coal mine feature data from the historical coal mine accident data and coal mine non-accident data is extracted as a coal mine risk prediction training set; A model building module is used to input the coal mine risk prediction training set into the logistic regression model for training to obtain a coal mine risk prediction model; Real-time data acquisition module, which obtains coal mine hidden danger text data and real-time data of measuring points; The risk assessment data acquisition module is used to extract the coal mine feature vectors from the coal mine hidden danger text data and the real-time data of the measuring points according to the coal mine feature extraction model as the coal mine risk assessment data; The risk prediction module is used to call the coal mine risk prediction model, perform risk prediction on the extracted coal mine risk assessment data, and obtain the coal mine risk prediction assessment results; The method for pre-building a coal mine feature extraction model includes the following sub-steps: Obtain hidden danger text vector training set; Divide the hidden danger text vector training set into a training set and a test set, use the training set to train the SVM classification model to obtain the text hidden danger classification model, use the test set to test the text hidden danger classification model, optimize the text hidden danger vector, and obtain the optimized text hidden danger vector; Obtaining the hidden danger vector of the measuring point, combining the hidden danger vector of the measuring point with the hidden danger vector of the text, and obtaining the coal mine feature vector; The method for obtaining a hidden danger text vector training set includes the following sub-steps: Obtain a manually annotated hidden danger text training set; Preprocess the hidden danger text training set; Performing vectorization processing on the preprocessed hidden danger text training set to obtain a hidden danger text vector training set; The method for vectorizing the pre-processed hidden danger text training set is as follows: using the TF-IDF calculation method to calculate the text data in the hidden danger text training set, thereby realizing the vectorization of the text data in the hidden danger text training set; TF-IDF is calculated as follows: TF-IDF=TF*IDF; Among them, TF represents the term frequency, which represents the frequency of the term t appearing in the document; IDF represents the inverse document frequency index. The main idea of ​​IDF is that if there are fewer documents containing the term t, that is, the number of times the keyword appears in the hidden danger text is smaller, the larger the IDF is, which means that the term t has a good ability to distinguish categories, TF i,j Indicates the frequency of the i-th word in the j-th document; n i,j Indicates the number of times the i-th word appears in the j-th document; ∑ k n k,j Represents the total number of times all words appear in the jth document; idf i represents the inverse text frequency index of the i-th word; D represents the document; |D| represents the total number of documents; |{j,t i ∈D i }| indicates that it contains the word t i the number of files; Among them, the method for obtaining the hidden danger vector of the measuring point is: Obtain real-time data from coal mine measurement points, including gas, oxygen, coal dust, wind speed, and mine pressure. Perform statistical judgment on the real-time data of the measuring points, and take the category corresponding to the risk hidden danger of the real-time data of the measuring points as the measuring point hidden danger vector.

Images