Intelligent fusion classification and grading early warning method for major disasters in coal mines

Abstract

The application discloses a kind of coal mine major disaster intelligent fusion classification grading early warning method.First, the data set is constructed by collecting coal mine safety monitoring data.Second, Anomaly-Transformer and E 2 GAN model are trained to identify and interpolate abnormal values and missing values in the data, achieving data preprocessing.Third, a new multi-index fusion prediction model is introduced by combining BiLSTM and MMOE multi-task learning network, which can predict the future trend of each index after training and optimization.Then, the AdaTT network is improved to realize the multi-disaster risk fusion early warning model, and the F1 index is used to optimize the number of training times and the early warning threshold.The real-time monitoring and fusion prediction data are input into the optimized early warning model to obtain the risk probability of each disaster.Finally, a multi-disaster grading early warning method is proposed by combining the best early warning threshold and risk probability, which realizes the grading identification of different disaster risks.This method can effectively warn the risks of gas, fire, dust, roof and other disasters, and provide risk information in advance, which is of great significance for the early control of coal mine multi-disaster risks.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of coal mine disaster early warning, in particular to a coal mine major disaster intelligent fusion classification grading early warning method. BACKGROUND

[0002] With the gradual depletion of shallow coal resources, the increase in mining depth leads to increasingly complex geological conditions. The risk of accidents such as gas, fire, dust, roof, water, rock burst, coal and gas outburst is increasingly serious, causing serious property loss and casualties. Comprehensive study of the mutual influence law between various indicators, the risk coupling relationship between various disasters and the different indicator risk information before the occurrence of various disaster accidents is an important prerequisite for the prediction and early warning of various disaster risks.

[0003] At present, the traditional early warning methods of disaster risks such as gas, coal dust, fire, roof, water, rock burst, coal and gas outburst are mostly based on the time sequence precursor characteristics and data range independent identification and early warning of single factor indicators, for example, trend method, critical value method or machine learning models for independent identification of each disaster, such as CNN, LSTM, etc. These methods ignore the mutual influence and correlation between various disasters and various indicators. Due to the complex coal mine disaster environmental conditions and coupling process, the influencing factors and evolution process are not clear. This early warning method considering only a single indicator cannot quantitatively and accurately express the disaster characteristics of various disasters, ignores the mutual influence relationship between various disasters and various indicators, lacks collaborative monitoring and application of multiple disasters and multiple indicators, and cannot accurately identify the danger of various disaster risks, leading to the risk and hidden danger cannot be identified and controlled in time, which is easy to cause the misreporting or misreporting of the identification of disaster risks such as gas, coal dust, fire, roof, water, rock burst, coal and gas outburst, reducing the accuracy and reliability of risk early warning. SUMMARY

[0004] The purpose of the present application is to provide a coal mine major disaster intelligent fusion classification grading early warning method, which can fuse factors such as gas, coal dust, fire, roof, water, rock burst, coal and gas outburst for early warning, and improve the accuracy and reliability of early warning.

[0005] Technical scheme: A coal mine major disaster intelligent fusion classification grading early warning method, comprising the following steps:

[0006] Step one: data preprocessing;

[0007] Collect a large amount of safety monitoring time series data in the coal mine, including time series data of gas concentration, carbon monoxide concentration, coal dust concentration, fully mechanized resistance, surrounding rock separation, anchor rod stress, temperature, wind speed, water inflow, etc. as initial data set, use deep learning model to identify and interpolate the abnormal data existing in the time series data, realize the preprocessing of data;

[0008] Step 2: Fusion Prediction;

[0009] A multi-indicator fusion prediction model is established based on a bidirectional long short-term memory network and an improved MMoE algorithm. The pre-processed data is input into the multi-indicator fusion prediction model. By learning the precursor trend characteristics of each indicator data and the correlation between indicators, the future development trend of different indicators can be obtained simultaneously.

[0010] Step 3: Integrated Early Warning;

[0011] A multi-hazard risk fusion early warning model based on the improved AdaTT multi-index fusion prediction model was established. The original data and prediction data were input into the fusion early warning model. The model was used to simultaneously output the current and future risk probabilities of each hazard by fusion analysis of the data precursor features and the relationship between hazard. The optimal model was determined by evaluation indicators, and the early warning threshold of each hazard risk was obtained by combining the accuracy distribution in the test set.

[0012] Step 4: Hierarchical identification;

[0013] By utilizing the current and future risk probabilities of each disaster obtained from the fusion early warning model, and combining the risk early warning thresholds of each disaster, a risk classification and early warning method for each disaster is established that integrates the current and future risk status. This method is used for comprehensive risk identification and early warning of disasters such as gas, fire, dust, roof collapse, water, rock bursts, and coal and gas outbursts.

[0014] Furthermore, in step one, the collected abnormal data is divided into training, validation, and test sets, and Anomaly-Transformer and E... 2 GANs build pre-trained models and train and test them, using optimized Anomaly-Transformer and E... 2 GAN identifies and imputes outliers and missing values ​​in time series data of gas concentration, carbon monoxide concentration, coal dust concentration, fully mechanized mining resistance, surrounding rock delamination, anchor cable stress, temperature, wind speed, and water inflow collected on-site.

[0015] Furthermore, in step two, the method for establishing a multi-indicator fusion prediction model includes:

[0016] Let x be the input data for all indicators, and let K = {k1, k2, ... k} be the indicator prediction task. n The number of expert networks is N. Long-term dependency features of each indicator are extracted using a bidirectional long short-term memory network. The calculation method is as follows:

[0017]

[0018] in, a candidate information memory cell of the kth index at the tth moment, an information memory cell of the kth index at the tth moment, information output of the kth index at the tth moment; then long-time dependence features learned by the bidirectional long short-term memory network of different indexes are spliced to obtain a splicing result X c :

[0019] X c =[H1,H2,...,H k ];

[0020] wherein H k is a result output by the bidirectional long short-term memory network of the kth index;

[0021] Finally, the spliced matrix is input into a multi-task learning model MMoE for fusion calculation, and the prediction results of each index are output synchronously after the task tower; the method is as follows:

[0022]

[0023] wherein G k (x) is a gating unit of the kth task, W gk is a learning parameter weight thereof, f k (x) is an output after learning of the kth task by a parameter-shared expert network, m k is a task tower network, and y k is a prediction result of the kth task obtained after fusion analysis of all index data.

[0024] Further, in step three, the method for determining each disaster warning threshold comprises:

[0025] Suppose that the input disaster risk monitoring index data is x, the warning task is T={t1,t2,…,t n}, the number of fusion unit stack layers is L={l1,l2,…,l n}, and the tth task fusion module in the lth layer is represented as The fusion output of the tth warning task in the lth layer is:

[0026]

[0027] The calculation process of each fusion network block FU is that firstly, the outputs of multiple expert networks of key indexes are integrated through a linear model, and the formula is as follows:

[0028]

[0029] wherein, is a matrix after superposition of each expert,​ is the parameter vector, NELF is the output of the key indicators after linear fusion by multiple experts; then, the relevance of other indicators and the key indicators is learned by a gating network, and the outputs of all expert networks are fused to obtain the influence weight of each factor on the disaster risk evolution, as follows:

[0030]

[0031] Finally, the gating network result obtained is added to the result of the fusion of the key indicator expert network, to obtain a data fusion block for the disaster risk warning task, as follows:

[0032]

[0033] Best, in step three, the optimal model of the warning model is determined by the optimal F1 value method, the model determination method based on the threshold of the maximum F1 value is used to obtain the iteration number of the model when the F1 is optimal in the iterative training process, and the threshold corresponding to the optimal model is determined according to the accuracy distribution in the test set. F1 is the harmonic mean of precision and recall. By adjusting the threshold, the values of precision and recall are further modified, so that the F1 value reaches the maximum, thereby determining the disaster risk warning threshold.

[0034] During the iteration process, due to the change of the threshold, the model obtained in one iteration can obtain multiple F1 values due to the change of the threshold. Here, based on an interval of 0.01, the maximum F1 is selected as the model of this iteration compared with other iteration numbers, and the corresponding threshold is equivalent to the warning threshold of the iteration model, that is, the F1 optimal threshold under different threshold conditions is selected for each iteration, which is calculated as the optimal F1 value of this iteration. The optimal model is selected from multiple iterations, and the best is selected from the best.

[0035] Best, the determination method of F1 value is:

[0036] Let the precision be P, and the precision P be the number of actual labels that are also warning in the model output as warning results, which accounts for the proportion of the number of model output as warning results, as shown in the following formula:

[0037]

[0038] Where TP represents the number of samples that are actually warned and predicted as warned, and FP represents the number of samples that are not actually warned but predicted as warned;

[0039] Let the recall be R, and the recall R be the proportion of the predicted value as warning and the true value as warning samples in all samples with true value as warning:

[0040]

[0041] Wherein, FN represents the number of samples actually warning but predicted as non-warning, the calculation method of F1 value is as follows:

[0042]

[0043] Further, in step four, the risk probability of gas, fire, coal dust, roof, water, rock burst, coal and gas outburst and other disasters at the current time point and the future time period can be obtained by using the fusion warning model, if the disaster risk probability does not exceed the threshold value in both time, no warning, if the disaster risk probability at the current time point exceeds the threshold value, and the disaster risk warning probability in the future time period also exceeds the threshold value, the corresponding disaster is warned at the first level, and the rest is warned at the second level.

[0044] The threshold value is comprehensively considered from the fusion warning F1 value and the distribution of the selection accuracy rate of the test set different threshold values.

[0045] Advantages: compared with the prior art, the advantages of the present application are: the present application combines deep learning with fusion warning and prediction of gas, fire, dust, roof, water, rock burst, coal and gas outburst and other disaster risks and fusion analysis of each monitoring index data, comprehensively considers the mutual influence law between each index, the risk coupling relationship between each disaster and the different index risk precursor characteristic information before each disaster accident, and performs early warning on the gas, coal dust, fire, roof, water, rock burst, coal and gas outburst and other disaster risks. The present application firstly pre-processes the original data sequence, then provides complete and high-quality sequence data for subsequent prediction and warning, then fusion predicts all indexes, the prediction model shares data information between multiple tasks through shared parameters, effectively learns the precursor trend characteristics of different index data and the association relationship with each prediction task, has high data prediction accuracy, and provides important future data support for subsequent multi-disaster risk fusion warning. Finally, the original data and future data are input into the fusion warning model at the same time, the fusion warning model comprehensively learns the association relationship between the precursor characteristics of different index historical data, different indexes and each disaster risk warning task, and different warning tasks, can improve the fusion analysis ability of the warning task, makes each disaster risk identification has high accuracy. In addition, the present application uses the fusion warning model to obtain the risk probability of the current and future gas, coal dust, fire, roof, water, rock burst, coal and gas outburst and other disaster risks, and performs graded warning on different disaster risks, better ensures the accurate identification of each disaster. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 The flowchart of the present application;

[0047] Figure 2 The data pre-processing training and application flowchart;

[0048] Figure 3 An improved MMoE multi-index fusion prediction model architecture diagram;

[0049] Figure 4 An improved AdaTT coal mine multi-disaster fusion early warning model architecture diagram. DETAILED DESCRIPTION

[0050] The present application will be further illustrated below in conjunction with the drawings and specific examples, and it should be understood that these examples are only used to illustrate the present application and not to limit the scope of the present application.

[0051] A coal mine major disaster intelligent fusion classification and grading early warning method, as shown in Figure 1 , comprising the following steps:

[0052] Step 1: data preprocessing;

[0053] The present application uses Anomaly-Transformer and E 2 GAN respectively to identify and interpolate the abnormal values and missing values existing in the time series data of gas concentration, carbon monoxide concentration, coal dust concentration, fully mechanized resistance, surrounding rock separation, anchor rod stress, temperature, wind speed, water inflow and the like collected on site. Anomaly-Transformer model realizes unsupervised time series anomaly value detection through abnormal attention mechanism and maximum minimum correlation strategy. The main idea of the abnormal attention mechanism is to use a learnable Gaussian kernel and a self-attention mechanism to learn the prior correlation of the features within the relative time and the sequence correlation characteristics of the original sequence, respectively, to obtain more information correlation when the sequence appears abnormal values, so that the abnormal score of each monitoring value in the time series can be calculated, and the model has good performance in identifying abnormal values. E 2 GAN model is an end-to-end generative adversarial network model for multivariate time series missing value estimation, mainly composed of a generator for complete sequence reconstruction of the low-dimensional vector mapped after adding noise to the original sequence and a discriminator for discriminating the reconstructed sequence and the real sequence. Its core is to guarantee the time lag phenomenon that the current sample is affected by the past samples and weakened with time through the GURI module, consider the influence of irregular sample missing on time series samples, and thus perform dimension reduction, reconstruction and discrimination of real samples and reconstructed samples on incomplete time series. After multiple rounds of iteration training of the generator and the discriminator, when the discriminator cannot discriminate between the original sample and the reconstructed sample, the corresponding value of the reconstructed sample can be used to fill in the missing values in the original sample. The model architecture, training and verification process are shown in Figure 2 .

[0054] Step 2: fusion prediction and early warning;

[0055] a) an improved MMoE-based deep learning fusion prediction model;

[0056] The present application is directed to gas, fire, dust, roof, water, rock burst, coal and gas outburst and other disaster monitoring indexes, and the future trend of each index can be more accurately predicted by comprehensively analyzing the coupling evolution law among them. The multi-task prediction model can simultaneously input continuous historical data (length n) of multiple index data, then fuse and analyze the correlation between indexes in the hidden layer, comprehensively consider the mutual influence between different indexes, and finally synchronously output the prediction data (length m) of all indexes. MMoE is a multi-task learning model with shared parameters as the base, and its basic idea is to learn the correlation of input data to different tasks through shared multiple expert networks and gating units, and then output the prediction values of different tasks through the task tower composed of feedforward neural networks. However, the original MMoE model is not good at extracting features of internal relations of long sequences, and adding a bidirectional long short-term memory network before the shared parameters can effectively obtain the evolution relationship between original sequences. Assuming that all input index data is x, the index prediction task is K={k1, k2, …k n}, and the number of expert networks is N, the long-time dependence features of each index are extracted through the bidirectional long short-term memory network, and the calculation method is as follows:

[0057]

[0058] Among them, is the candidate information memory cell of k index at t moment, is the information memory cell of k index at t moment, is the information output of k index at t moment; then the long-time dependence features learned by different indexes through the bidirectional long short-term memory network are spliced to obtain the splicing result X c :

[0059] X c =[H1,H2,…,H k ];

[0060] Among them, H k is the output result of the kth index through the bidirectional long short-term memory network;

[0061] Finally, the spliced matrix is input into the multi-task learning model MMoE for fusion calculation, and the prediction results of each index are synchronously output after the task tower. The fusion calculation method of MMoE is as follows:

[0062]

[0063] Among them, G k (x) is the gating unit of k task, W gkLearning parameters weights for it. k (x) is the output of the k task learned by the parameter sharing expert network, m k is the task tower network, y k is the k task prediction result obtained by analyzing all index data fusion. The improved model architecture is as follows Figure 3 .

[0064] b) deep learning fusion prediction model based on improved AdaTT;

[0065] The fusion early warning model synchronously inputs multiple index data, uses the data fusion module and the multi-disaster risk early warning task fusion module to learn the mutual correlation between the precursor characteristics of the historical data of each index, each index and each disaster risk early warning, and the evolution process between each disaster risk, and synchronously outputs the fusion early warning result of different disasters. However, different indexes have different response degrees to each disaster risk, therefore, the present application selects the working face gas concentration, carbon monoxide concentration, dust concentration, fully mechanized resistance, water inflow as the key factors for disaster risk early warning of gas, fire, coal dust, roof, water, rock burst, coal and gas outburst, etc., and the surrounding rock separation, anchor rod stress, temperature, wind speed, long observation hole, water level as auxiliary factors. Adatt is a multi-task adaptive deep fusion network model, which is constructed by specific early warning tasks and optional shared fusion blocks at multiple levels. The main idea is to learn the knowledge of specific index data rules by assigning different expert networks to different early warning tasks, and to perform data-to-data and task-to-task fusion by using residual mechanism and gating mechanism. These units adaptively learn the associated knowledge and rules between different data and specific early warning tasks, thereby effectively modeling the relationship between each early warning task and realizing knowledge fusion and sharing. By stacking multiple fusion modules, the early warning result of a specific early warning task under the influence of different index historical data precursor characteristics and other early warning tasks can be learned from a large amount of data.

[0066] Assuming that the input disaster risk monitoring index data is x, the early warning task is T={t1, t2, …, t n}, the number of fusion unit stacking layers is L={l1, l2, …, l n}, and the tth task fusion module in the lth layer is represented as The fusion output of the tth early warning task in the lth layer is .

[0067]

[0068] The calculation process of each fusion network block FU is to first integrate the outputs of multiple expert networks of key indexes through a linear model, and the formula is as follows:

[0069]

[0070] wherein is the superimposed matrix of each expert, is the parameter vector, and NELF is the output of the key indicator after linear fusion by multiple experts. Then, the relevance of other indicators to the key indicator is learned by using a gating network, and the outputs of all expert networks are fused to obtain the influence weight of each factor on disaster risk evolution, as follows:

[0071]

[0072] Finally, the gating network result and the fusion result of the key indicator expert network are added to obtain a data fusion block for the disaster risk warning task, as follows:

[0073]

[0074] The overall architecture of the model is shown in Figure 4 .

[0075] In order to better select the field application model, the present application uses a threshold model determination method based on F1 value maximization to obtain the F1 optimal model iteration number in the iterative training process and determine the corresponding threshold value of the optimal model in the test set. F1 is the harmonic mean of precision and recall, and by adjusting the threshold value, the values of precision and recall are further modified to maximize the F1 value, thereby determining the disaster risk warning threshold value. The precision P refers to the number of samples that are actually labeled as warning and are output as warning by the model, which accounts for the proportion of the number of samples that are output as warning by the model, and is calculated as follows:

[0076]

[0077] wherein TP represents the number of samples that are actually warning and are predicted as warning, and FP represents the number of samples that are actually not warning but are predicted as warning. The recall R refers to the proportion of samples that are predicted as warning and are actually warning in all samples that are actually warning.

[0078]

[0079] wherein FN represents the number of samples that are actually warning but are predicted as not warning, and the calculation method of the F1 value is as follows.

[0080]

[0081] Step three: hierarchical identification and warning

[0082] The hierarchical identification early warning is carried out by distributing the current risk probability and the future risk probability. Specifically, the risk probability of disasters such as gas, fire, coal dust, roof, water, rock burst, coal and gas outburst at the current time point and the future time period can be obtained by using the fusion early warning model. If the disaster risk probability does not exceed the threshold value in both time periods, no early warning is given. If the disaster risk probability at the current time point exceeds the threshold value, and the disaster risk early warning probability in the future time period also exceeds the threshold value, the corresponding disaster is given a first-level early warning, and the rest is given a second-level early warning.

[0083] Firstly, a large amount of safety monitoring time series data in different coal mines is collected as the initial training data set. Secondly, Anomaly-Transformer and E 2 The GAN model identifies and supplements the abnormal data such as abnormal values and missing values in the time series data, and realizes the preprocessing of the data. Then, a new multi-index fusion prediction model is introduced by combining the bidirectional long short-term memory network and the MMOE multi-task learning network. The preprocessed index data is input into the trained and optimized fusion prediction model, and the future development trend of each index can be obtained synchronously. Next, the AdaTT model is improved to realize the multi-disaster risk fusion early warning model based on the multi-task learning network. The F1 evaluation index is combined to obtain the model training iteration number and the best risk early warning threshold value of each disaster when the F1 value is optimal. The real-time data and the fusion prediction data of each monitoring index are input into the fusion early warning model trained and optimized by the field data, and the risk probability of each disaster at the current time point and the future time period can be obtained synchronously. Finally, a multi-disaster hierarchical identification early warning method is proposed by combining the best risk early warning threshold value and the risk probability obtained by the fusion early warning model. The risk early warning status of different disasters in the current and future time is comprehensively evaluated to realize the risk hierarchical identification early warning of each disaster. The present application constructs an intelligent fusion classification and hierarchical early warning method for major disasters in coal mines, mainly including data preprocessing, multi-index fusion prediction, multi-disaster fusion early warning and hierarchical identification. The method can effectively classify and respond to the risk of disasters such as gas, fire, dust, roof, water, rock burst, coal and gas outburst, and can provide risk information in advance, which is of great significance for accurate early warning of coal mine multi-disaster risk.

Claims

1. A coal mine major disaster intelligent fusion classification grading early warning method, characterized in that The method comprises the following steps: Step one: data preprocessing; Collecting a large amount of safety monitoring time series data in the coal mine, including time series data of gas concentration, carbon monoxide concentration, coal dust concentration, fully mechanized resistance, surrounding rock separation, anchor cable stress, temperature, wind speed and water inflow as an initial data set, using a deep learning model to identify and interpolate abnormal data in the time series data, and realizing data preprocessing; Step two: fusion prediction; A multi-index fusion prediction model is established based on a bidirectional long short-term memory network combined with an improved MMoE algorithm, the preprocessed data is input into the multi-index fusion prediction model, the correlation between the indexes is learned, and the future development trend of different indexes is obtained synchronously; In step two, the method for establishing the multi-index fusion prediction model comprises: Let all the input index data be X = {x1, x2, …, x t}, the index prediction task be K = {k1, k2, … k n}, and the number of expert networks be N. The long-time dependence features of each index are extracted by a bidirectional long short-term memory network, and the calculation method is as follows: wherein, is the candidate information memory cell of the k index at the t moment, is the information memory cell of the k index at the t moment, is the information output of the k index at the t moment; then the long-time dependence features learned by the bidirectional long short-term memory network of different indexes are spliced to obtain a splicing result X c : X c = [H1, H2,..., H k ]; where H k is the result of the kth index output by the bidirectional LSTM network; Finally, the spliced matrix is input into the multi-task learning model MMoE for fusion calculation, and the prediction results of each index are output synchronously after the task tower; the method is as follows: G k (x) = softmax(W gk x) y k = m k (f k (x)) Wherein, G k (x) is the gating unit of k tasks, W gk is the learning parameter weight thereof, f k (x) is the output after learning of the expert network of k tasks through parameter sharing, m k is the task tower network, y k is the prediction result of k tasks obtained through fusion analysis of all index data; Step three: fusion early warning; A multi-disaster risk fusion early warning model based on the multi-index fusion prediction model of the improved AdaTT is established, the original data and the prediction data are input into the fusion early warning model, the precursor characteristics of the historical and future data and the fusion analysis and identification of the relationship between disasters are used, and the current and future risk probabilities of each disaster are output synchronously; the optimal model is determined through the evaluation index, and the risk early warning threshold of each disaster is obtained according to the accuracy distribution in the test set; Step four: hierarchical identification; The current and future risk probabilities of each disaster obtained by the fusion early warning model are used to identify and warn the comprehensive risk of gas, fire, dust, roof, water, rock burst and coal and gas outburst disasters in combination with the risk early warning threshold of each disaster.

2. The intelligent fusion classification grading early warning method for major disasters in coal mines according to claim 1, characterized in that, In step one, the collected abnormal data is divided into training set, validation set and test set, Anomaly-Transformer and E 2 GAN builds data pre-training model and trains and tests, and identifies and fills in the existing abnormal values and missing values in the time series data of gas concentration, carbon monoxide concentration, coal dust concentration, fully mechanized resistance, surrounding rock separation, anchor cable stress, temperature, wind speed and water inflow collected on site. 2 GAN respectively identifies and fills in the existing abnormal values and missing values in the time series data of gas concentration, carbon monoxide concentration, coal dust concentration, fully mechanized resistance, surrounding rock separation, anchor cable stress, temperature, wind speed and water inflow collected on site.

3. The intelligent fusion classification grading early warning method for major disasters in coal mines according to claim 1, characterized in that, In step three, the optimal model of the early warning model is determined by the optimal F1 value method, the threshold model determination method based on F1 value maximization is used to obtain the iteration number of the optimal F1 value in the iteration training process, and the threshold corresponding to the optimal model in the test set is determined according to the accuracy distribution, F1 is the harmonic mean of precision and recall, the values of precision and recall are further modified by adjusting the threshold, so that the F1 value reaches the maximum, and thus the risk early warning threshold of each disaster is determined.

4. The intelligent fusion classification grading early warning method for major disasters in coal mines according to claim 3, characterized in that, The determination method of F1 value is as follows: Let the precision be P, the precision P is the number of actual labels that are also prewarned in the prewarned results output by the model, which accounts for the proportion of the number of prewarned results output by the model, and is calculated as shown in the following formula: Where TP represents the number of samples that are actually prewarned and predicted as prewarned, and FP represents the number of samples that are actually not prewarned but predicted as prewarned; Let the recall be R, the recall R is the proportion of samples whose predicted value is prewarned and whose true value is also prewarned in all samples whose true value is prewarned: Where FN represents the number of samples that are actually prewarned but predicted as not prewarned, and the calculation method of F1 value is as follows:

5. The intelligent fusion classification grading early warning method for major disasters in coal mines according to claim 1, characterized in that, In step four, the risk probabilities of gas, fire, coal dust, roof, water, rock burst, and coal and gas outburst disasters are calculated by using the fusion early warning model on the precursor characteristics of historical and future data. If the disaster risk probabilities in both time periods do not exceed the threshold value, no early warning is given. If the disaster risk probability in the current time point exceeds the threshold value and the disaster risk warning probability in the future time period also exceeds the threshold value, a corresponding disaster is given a first-level early warning. Otherwise, a second-level early warning is given.

Images