Mine fire grading early warning method based on intelligent group algorithm

Abstract

The invention provides a mine fire grading early warning method based on an intelligent group algorithm, and the method comprises the steps: constructing sample data sets of different fire risk grades based on data indexes of different sampling frequencies collected by a coal mine underground upper corner monitoring system and a goaf beam tube; an attention mechanism is introduced, and an extreme learning machine fire early warning model with the upper corner CO concentration index as the dominant weight is constructed; inputting the training set into an input layer of an extreme learning machine fire early warning model, and optimizing monitoring index residues of the extreme learning machine fire early warning model in real time in combination with a grey wolf intelligent group algorithm to obtain an optimized fire grading early warning model; and inputting an upper corner monitoring system and a goaf beam tube monitoring index as a test set into the optimized fire hazard grading early warning model to obtain a coal mine underground fire hazard risk grade condition. The mine fire early warning level can be improved, and safe production of a coal mine is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine disaster monitoring and early warning, and in particular to a mine fire classification early warning method based on an intelligent swarm algorithm. Background Art

[0002] Coal will remain the main energy source in my country for a long time in the future. Mine fire is one of the disasters that pose a major threat to mine safety production, and is often accompanied by coal spontaneous combustion disasters in mine production. The main areas of coal spontaneous combustion danger include working face goaf, closed goaf, tunnel coal pillars, etc. According to the different heat sources of ignition, mine fires are usually divided into two categories: internal fire and external fire. Whether it is an internal or external cause fire, before the fire occurs, it will be accompanied by abnormal changes in characteristic gases and environmental monitoring indicators, such as CO, C2H2, C2H4, O2, wind speed, wind direction, temperature, and humidity. The degree of change in these monitoring data can directly reflect the fire risk situation, especially the CO concentration monitored in the upper corner can significantly reflect the risk of coal spontaneous combustion, and the C2H2 and C2H4 monitored by the goaf bundle tube have a certain lag, which can be used as auxiliary indicators for fire warning. The current monitoring indicators for mine fire early warning are not fully utilized, and there is no comprehensive consideration of different indicator data indicators in the upper corner and goaf. The early warning results are single, and the lack of intelligent optimization algorithm support leads to insufficient accuracy of early warning based on the neural network model. Therefore, it is necessary to propose an intelligent advanced method for fire graded early warning based on multiple monitoring indicators in coal mines to solve the above problems.

[0003] The Chinese invention patent, with publication number "CN109345754A", is named "A mine fire early warning monitoring system and method based on BP neural network", which discloses a mine fire early warning monitoring system and method based on BP neural network, including a sensor module, a server, a controller, a start module, a fire extinguishing system, an alarm module, a central control computer, and analysis software. Among them: the sensor module is connected to the server; the server is connected to the central control computer and the controller respectively; the controller is connected to the start module and the alarm module respectively; the start module is connected to the fire extinguishing system to start the fire extinguishing system; the central control computer is connected to the server and is equipped with analysis software to receive the parameters transmitted by the server. This technical solution can monitor the data of the fire-prone areas in the mine in real time, find the fire point in time, and take feedback measures. However, although this technical solution improves the fire warning capability and adaptability to large amounts of monitoring data by introducing the BP neural network, the research mainly focuses on the development of software and hardware modules and system architecture. The research on the early warning model method is not in-depth enough, and the different data indicators of the upper corner and goaf are not comprehensively considered. The early warning results are single, and the lack of support from intelligent optimization algorithms leads to insufficient accuracy of early warnings based on neural network models. Summary of the invention

[0004] In order to solve the problem that the above-mentioned prior art does not fully utilize monitoring indicators for mine fire warning, does not comprehensively consider different indicator data indicators of upper corners and goafs, and the warning results are single, lacking the support of intelligent optimization algorithms, resulting in insufficient accuracy of warnings based on neural network models. The present invention proposes a graded warning method for mine fires based on an intelligent swarm algorithm. The method is based on fire monitoring data of different frequencies collected by the upper corner monitoring system of the mine and the goaf bundle tube, including CO, C2H2, C2H4, O2, wind speed, wind direction, temperature, and humidity. The ELM neural network model is combined with the gray wolf swarm optimization algorithm to determine the mine fire risk level, so as to achieve graded warning of underground coal mine fires.

[0005] The present invention is achieved through the following technical solution: comprising the following steps:

[0006] The mine fire graded early warning method based on intelligent swarm algorithm is characterized by comprising the following steps:

[0007] Step 1: Obtain the CO, O2, wind speed, wind direction, temperature, humidity collected by the upper corner monitoring system and the C2H2, C2H4 index data collected by the goaf bundle pipe;

[0008] Step 2: Perform normalization and upsampling preprocessing on all indicator data in step 1, create sample data sets of different risk levels, and divide them into training sets and test sets;

[0009] Step 3: Introduce the attention mechanism and construct an extreme learning machine fire warning model with the corner CO concentration index as the dominant weight. The formula is as follows:

[0010]

[0011] η j =[η j1 ,η j2 ,...,η jn ] T

[0012]

[0013] The normalized CO, O2, wind speed, wind direction, temperature, humidity collected by the upper corner monitoring system and the C2H2, C2H4 index data collected by the goaf bundle pipe in step 2 are used as E i 'Network input; is the connection weight between the input layer and the jth node in the hidden layer; η j is the output weight connecting the jth hidden layer node to the output node; j is the threshold of the jth hidden layer node, is the connection weight between the input layer and the jth node in the hidden layer and E i 'The inner product of the network input; m is the number of hidden layer nodes, n is the number of output layer nodes; f(·) is the activation function; four risk levels of no, weak, medium and strong are used as y i Early warning network model output, R i It is a monitoring data indicator that is weighted by importance through the attention mechanism. is the weight coefficient of the importance of the monitoring indicator;

[0014] Step 4: Use the Gray Wolf Intelligent Swarm Algorithm to optimize the network parameters of the extreme learning machine fire warning model obtained in step 3, thereby obtaining an optimized fire graded warning model;

[0015] Step 5: Input the real-time data of CO, C2H2, C2H4, O2, wind speed, wind direction, temperature, and humidity monitored underground into the optimized fire classification warning model to obtain the fire classification warning results;

[0016] Step 6: According to the fire classification warning results, determine whether there is a fire hazard and the risk level; use accuracy (Acc), precision (Precison), recall (Recall), and F1 score as evaluation indicators to evaluate the warning effect of the model.

[0017] As further preferred, step 4 includes the following specific steps:

[0018] Step 4-1: Set the initialization parameters of the gray wolf intelligent group algorithm, including the number of wolves in the wolf pack M, the maximum number of iterations T, the maximum expected error value ε, and the length d of the position vector of the weight and threshold;

[0019] Where d = p + q, p is the weight of the neural network of the extreme learning machine fire warning model The number of, q is the number of thresholds h;

[0020] Step 4-2: According to formula X i = l h +g×(u h -l h ), get the initial wolf pack position vector;

[0021] The individual position vector X of the gray wolf i Corresponding weights in the extreme learning machine fire warning model and threshold h,X i The length of the individual code is the sum of the weight and the threshold number, g is a random number vector; u h and l h are the upper and lower bounds of the wolf pack’s location, respectively;

[0022] Step 4-3: According to the formula Get the fitness value fit;

[0023] Where: i is the number of nodes in the output layer; Y i is the actual fire danger level output value of node i, Z i is the predicted output value of the fire risk level of node i;

[0024] Step 4-4: According to the fitness value fit obtained in step 4-3, the position vectors of wolf α, wolf β, and wolf δ and the information of individual wolves moving toward the prey in the next iteration are obtained;

[0025] Calculate the information of the three gray wolves with the best fitness value fit from the first iteration to now, which are represented as: G α =fit best ; G β =fit second ; G δ =fit third ; Where: fit best 、fit second 、fit third They represent the three highest fitness values ​​of this iteration;

[0026] Update the position vectors of the three gray wolves α, β, and δ corresponding to the best fitness value fit: X α =X best ;X β =X second ;X δ =X third ;X best , X second , X third Fit best 、fit second 、fit third The corresponding position vector;

[0027] The formula of the fire warning model of the Gray Wolf Intelligent Group Algorithm Iterative Training Extreme Learning Machine is as follows:

[0028]

[0029] X(t)=[X1(t+1)+X2(t+1)+X3(t+1)] / 3 (4)

[0030] Where D is the distance between the wolf and its prey; X P (*) represents the position vector of the prey; X(*) represents the individual position vector of the gray wolf; t and T are the current and maximum iteration times, respectively; A and C are the group coordination coefficient vectors; r1, r2 are random numbers;

[0031] The distance between individual wolves and α, β, δ is determined by equations (2) and (3), and finally the information of individual wolves moving toward prey in the next iteration is calculated by equation (4); then, equation (2) is substituted into equation (3) to obtain the new position vector of the wolf pack in the next iteration;

[0032] Step 4-5: Determine whether the optimization conditions are met. If the number of iterations has not reached T, and the fitness value fit error of two consecutive trainings is not less than the maximum expected error value ε, return to step 4-2; otherwise, output the optimal gray wolf position vector X α , execute steps 4-6;

[0033] Step 4-6: Output the optimal gray wolf position vector X according to step 4-5 α , and obtain the corresponding weights in the extreme learning machine fire warning model and threshold h, and construct an optimized fire classification warning model.

[0034] As a further preferred embodiment, the calculation formula of the evaluation model in step 6 is as follows:

[0035]

[0036] In the formula, TP is a true positive example, which is a positive sample predicted as positive by the model; FP is a false positive example, which is a negative sample predicted as positive by the model; FN is a false negative example, which is a positive sample predicted as negative by the model; TN is a true negative example, which is a negative sample predicted as negative by the model.

[0037] As a further preferred embodiment, the formula for normalizing all indicator data in step 2 is as follows:

[0038]

[0039] In the formula, E i ' is the normalized data, E i is the original data, E max and E min are the maximum and minimum values ​​in the sample data set, respectively.

[0040] As a further preference, the value range of g in step 4-2 is [0,1].

[0041] As a further preference, in step 4-4, the value range of r1 and r2 is [0, 1].

[0042] As a further preference, in step 2, the division ratio of the training set to the test set is 7:3.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. The present invention performs normalization preprocessing on the CO, O2, wind speed, wind direction, temperature, humidity collected by the upper corner monitoring system and the C2H2, C2H4 indicator data monitored by the goaf bundle tube, thereby avoiding the influence of different monitoring data sampling rates and amplitude ranges on the results.

[0045] 2. The present invention introduces an attention mechanism to construct an extreme learning machine fire warning model with CO concentration as the dominant weight; determines the number of input layer units and hidden layer units of the extreme learning machine fire warning model according to the step size of the sample data; inputs the training set into the input layer unit of the extreme learning machine fire warning model, and uses the Grey Wolf intelligent swarm algorithm to train the parameters of the extreme learning machine fire warning model to obtain an optimized fire classification warning model; further, the coal mine underground fire monitoring data is input into the trained optimized fire classification warning model based on the Grey Wolf intelligent swarm algorithm and the extreme learning machine neural network to judge the coal mine underground fire risk level, thereby improving the accuracy of the coal mine underground fire classification warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0047] Advantages and features of the present invention will be illustrated and explained by the following non-limiting description of preferred embodiments thereof, which are given by way of example only with reference to the accompanying drawings.

[0048] like Figure 1 As shown, the present invention proposes a graded early warning method for mine fires based on an intelligent swarm algorithm. Different from the existing monitoring and early warning method, this method is based on a variety of mine fire index data from the upper corner monitoring system and goaf bundle pipe monitoring, and uses the Gray Wolf Intelligent Swarm Algorithm (GWO) combined with the Extreme Learning Machine (ELM) neural network to determine the mine fire risk level, obtain the optimal over-limit threshold of the fire monitoring index, and realize graded early warning of underground coal mine fire risks. It includes the following steps:

[0049] Step 1: Obtain the CO, O2, wind speed, wind direction, temperature, humidity collected by the upper corner monitoring system and the C2H2, C2H4 index data collected by the goaf bundle pipe;

[0050] Step 2: For all indicator data in step 1, use the formula Normalization and upsampling preprocessing are performed to create sample data sets of different risk levels, and the training set and test set are divided into a ratio of 7:3.

[0051] Due to the differences in sampling frequency and amplitude range between the monitoring system and the beam tube, normalization and upsampling preprocessing are performed to produce sample data sets of different risk levels, and the training set and test set are divided into a ratio of 7:3. The normalization formula is calculated as follows:

[0052]

[0053] In the formula, E i ' is the normalized data, E i is the original data, E max and E min are the maximum and minimum values ​​in the sample data set, respectively.

[0054] Step 3: Introduce the attention mechanism and construct an extreme learning machine (ELM) fire warning model with the upper corner CO concentration index as the dominant weight;

[0055] The training set sample CO, C2H2, C2H4, O2, wind speed, wind direction, temperature, and humidity monitoring data are input into the input layer of the fire classification warning model. The output value of the fire risk level of the network model is obtained by the inner product calculation with the weight and the output of the activation function. The calculation formula is as follows:

[0056]

[0057] The normalized CO, O2, wind speed, wind direction, temperature, humidity collected by the upper corner monitoring system and the C2H2, C2H4 index data collected by the goaf bundle pipe in step 2 are used as E i 'Network input; is the connection weight between the input layer and the jth node in the hidden layer; η j is the output weight connecting the jth hidden layer node to the output node; j is the threshold of the jth hidden layer node, is the connection weight between the input layer and the jth node in the hidden layer and E i 'The inner product of the network input; m is the number of hidden layer nodes, n is the number of output layer nodes; f(·) is the activation function; four risk levels of no, weak, medium and strong are used as y i Early warning network model output, R i It is a monitoring data indicator that is weighted by importance through the attention mechanism. It is the weight coefficient of the importance of the monitoring indicator.

[0058] Step 4: Use the Gray Wolf Intelligent Swarm Algorithm to optimize the network parameters of the extreme learning machine fire warning model obtained in step 3, thereby obtaining an optimized fire graded warning model;

[0059] Step 4-1: Set the initialization parameters of the gray wolf intelligent group algorithm, including the number of wolves in the wolf pack M, the maximum number of iterations T, the maximum expected error value ε, and the length d of the position vector of the weight and threshold.

[0060] Set the initialization parameters of the Gray Wolf Intelligent Swarm Algorithm for optimizing the extreme learning machine fire warning model, including:

[0061] (1) The number of wolves in the pack is M

[0062] (2) Maximum number of iterations T

[0063] (3) Training termination condition: The present invention adopts the maximum expected error value ε, that is, when the maximum number of iterations T is not reached, if the error between two adjacent training results is less than ε, the training ends and the optimized early warning model is output;

[0064] (4) Determine the length d of the position vector of the weight and threshold to form a 1×d dimensional matrix. Among them, the weight of the neural network of the extreme learning machine fire warning model is The number of is p, the number of thresholds h is q, d is the length of the individual position vector in the gray wolf intelligent group algorithm, d = p + q.

[0065] Step 4-2: According to formula X i = l h +g×(u h -l h ) to obtain the initial wolf pack position vector.

[0066] Where X i is the position vector of each gray wolf, i = 1, 2, ..., M, which contains all the weights of the neural network of the extreme learning machine fire warning model and thresholds (h1,h2,...,h q ), the individual position vector X of the gray wolf i Corresponding weights in the extreme learning machine fire warning model and threshold h,X i The length of the individual code is the sum of the weight and the threshold number, g is a random number vector between [0,1]; u h and l h are the upper and lower bounds of the wolf pack’s location, respectively.

[0067] Step 4-3: According to the formula Get the fitness value fit;

[0068] The fitness function fit(·) is determined by the sum of the absolute values ​​of the errors between the predicted value and the actual output value. The formula is as follows:

[0069]

[0070] Where: i is the number of nodes in the output layer; Y i is the actual fire danger level output value of node i, Z i is the predicted output value of the fire risk level of node i.

[0071] Step 4-4: According to the fitness value fit obtained in step 4-3, the position vectors of wolf α, wolf β, and wolf δ are obtained, as well as the information of the movement of individual wolves toward the prey in the next iteration.

[0072] Calculate the information of the three gray wolves with the best fitness value fit from the first iteration to now, which are represented as: G α =fit best ; G β =fit second ; G δ =fit third ; Where: fit best 、fit second 、fit third They represent the three highest fitness values ​​in this iteration.

[0073] Update the position vectors of the three gray wolves α, β, and δ corresponding to the best fitness value fit: X α =X best ;X β =X second ;X δ =X third ;X best , X second , X third Fit best 、fit second 、fit third The corresponding position vector.

[0074] The process of iterative training of the extreme learning machine fire warning model by the Gray Wolf intelligent swarm algorithm can be expressed as:

[0075]

[0076] Where D is the distance between the wolf and its prey; X P (*) represents the position vector of the prey; X(*) represents the individual position vector of the gray wolf; t and T are the current and maximum iteration times, respectively; A and C are the group coordination coefficient vectors; r1, r2 are random numbers between [0,1].

[0077]

[0078] X(t)=[X1(t+1)+X2(t+1)+X3(t+1)] / 3 (6)

[0079] The distance between individual wolves and α, β, and δ is determined by equations (4) and (5), and finally, equation (6) is used to calculate the information of individual wolves moving toward prey in the next iteration. Then, equation (4) is substituted into equation (5) to obtain the new position vector of the wolf pack in the next iteration.

[0080] Step 4-5: Determine whether the optimization conditions are met. If the number of iterations has not reached T, and the fitness value fit error of two consecutive trainings is not less than the maximum expected error value ε, return to step 4-2; otherwise, output the optimal gray wolf position vector X α , perform steps 4-6.

[0081] Step 4-6: Output the optimal gray wolf position vector X according to step 4-5 α , and obtain the corresponding weights in the extreme learning machine fire warning model and threshold h, and construct an optimized fire classification warning model.

[0082] According to the position vector X of wolf α in the gray wolf intelligent group algorithm α , determine the optimized weights and thresholds (h1,h2,...,h q ), and assign it to the neural network of the extreme learning machine fire warning model, based on which the optimized fire classification warning model is constructed.

[0083] Step 5: Input the real-time data of CO, C2H2, C2H4, O2, wind speed, wind direction, temperature, and humidity monitored underground into the optimized fire classification warning model to obtain the fire classification warning results.

[0084] Step 6: According to the fire classification warning results, determine whether there is a fire hazard and the risk level. Accuracy (Acc), precision (Precison), recall (Recall), and F1 score are used as evaluation indicators to evaluate the warning effect of the model. The calculation formula is as follows:

[0085]

[0086] In the formula, TP is a true positive sample predicted by the model as positive; FP is a false positive sample predicted by the model as positive; FN is a false negative sample predicted by the model as negative; TN is a true negative sample predicted by the model as negative.

[0087] The above-mentioned mine fire graded warning method based on intelligent swarm algorithm obtains a data sample set through the CO, O2, wind speed, wind direction, temperature, humidity and C2H2, C2H4 indicator data collected by the upper corner monitoring system and the goaf bundle pipe monitoring. Due to the different sampling rates and amplitude ranges of the monitoring data, resampling and normalization preprocessing are performed, and the data sample set is divided into a training set and a test set in a ratio of 7:3; an attention mechanism is introduced to construct an extreme learning machine fire warning model with CO concentration as the dominant weight; the number of input layer units and hidden layer units of the extreme learning machine fire warning model is determined according to the step size of the sample data; the training set is input into the input layer unit of the extreme learning machine fire warning model, and the parameters of the extreme learning machine fire warning model are trained using the Gray Wolf intelligent swarm algorithm to obtain an optimized fire graded warning model; further, the coal mine underground fire monitoring data is input into the trained optimized fire graded warning model based on the Gray Wolf intelligent swarm algorithm and the extreme learning machine neural network to judge the coal mine underground fire risk level, thereby improving the accuracy of the coal mine underground fire graded warning.

[0088] In addition to the above embodiments, the present invention may also have other implementation modes. Any technical solutions formed by equivalent replacement or equivalent transformation shall fall within the protection scope required by the present invention.

Claims

1. A mine fire graded early warning method based on intelligent swarm algorithm, characterized by: The steps include: Step 1: Obtain the CO, O2, wind speed, wind direction, temperature, humidity collected by the upper corner monitoring system and the C2H2, C2H4 index data collected by the goaf bundle pipe; Step 2: Perform normalization and upsampling preprocessing on all indicator data in step 1, create sample data sets of different risk levels, and divide them into training sets and test sets; Step 3: Introduce the attention mechanism and construct an extreme learning machine fire warning model with the corner CO concentration index as the dominant weight. The formula is as follows: The normalized CO, O2, wind speed, wind direction, temperature, humidity collected by the upper corner monitoring system and the C2H2, C2H4 index data collected by the goaf bundle pipe in step 2 are used as E i 'Network input; is the connection weight between the input layer and the jth node in the hidden layer; η j is the output weight connecting the jth hidden layer node to the output node; j is the threshold of the jth hidden layer node, is the connection weight between the input layer and the jth node in the hidden layer and E i 'The inner product of the network input; m is the number of hidden layer nodes, n is the number of output layer nodes; f(·) is the activation function; four risk levels of no, weak, medium and strong are used as y i Early warning network model output, R i It is a monitoring data indicator that is weighted by importance through the attention mechanism. is the weight coefficient of the importance of the monitoring indicator; Step 4: Use the Gray Wolf Intelligent Swarm Algorithm to optimize the network parameters of the extreme learning machine fire warning model obtained in step 3, thereby obtaining an optimized fire graded warning model; Step 5: Input the real-time data of CO, C2H2, C2H4, O2, wind speed, wind direction, temperature, and humidity monitored underground into the optimized fire classification warning model to obtain the fire classification warning results; Step 6: According to the fire classification warning results, determine whether there is a fire hazard and the risk level; use accuracy (Acc), precision (Precison), recall (Recall), and F1 score as evaluation indicators to evaluate the warning effect of the model.

2. The mine fire graded early warning method based on intelligent swarm algorithm according to claim 1 is characterized by: The step 4 comprises the following specific steps: Step 4-1: Set the initialization parameters of the gray wolf intelligent group algorithm, including the number of wolves in the wolf pack M, the maximum number of iterations T, the maximum expected error value ε, and the length d of the position vector of the weight and threshold; Where d = p + q, p is the weight of the neural network of the extreme learning machine fire warning model The number of, q is the number of thresholds h; Step 4-2: According to formula X i = l h +g×(u h -l h ), get the initial wolf pack position vector; The individual position vector X of the gray wolf i Corresponding weights in the extreme learning machine fire warning model and threshold h,X i The length of the individual code is the sum of the weight and the threshold number, g is a random number vector; u h and l h are the upper and lower bounds of the wolf pack’s location, respectively; Step 4-3: According to the formula Get the fitness value fit; Where: i is the number of nodes in the output layer; Y i is the actual fire danger level output value of node i, Z i is the predicted output value of the fire risk level of node i; Step 4-4: According to the fitness value fit obtained in step 4-3, the position vectors of wolf α, wolf β, and wolf δ and the information of individual wolves moving toward the prey in the next iteration are obtained; Calculate the information of the three gray wolves with the best fitness value fit from the first iteration to now, which are represented as: G α =fit best ; G β =fit second ; G δ =fit third ; Where: fit best 、fit second 、fit third They represent the three highest fitness values ​​of this iteration; Update the position vectors of the three gray wolves α, β, and δ corresponding to the best fitness value fit: X α =X best ;X β =X second ;X δ =X third ;X best , X second , X third Fit best 、fit second 、fit third The corresponding position vector; The formula of the fire warning model of the Gray Wolf Intelligent Group Algorithm Iterative Training Extreme Learning Machine is as follows: X(t)=[X1(t+1)+X2(t+1)+X3(t+1)] / 3 (4) Where D is the distance between the wolf and its prey; X P (*) represents the position vector of the prey; X(*) represents the individual position vector of the gray wolf; t and T are the current and maximum iteration times, respectively; A and C are the group coordination coefficient vectors; r1, r2 are random numbers; The distance between individual wolves and α, β, δ is determined by equations (2) and (3), and finally the information of individual wolves moving toward prey in the next iteration is calculated by equation (4); then, equation (2) is substituted into equation (3) to obtain the new position vector of the wolf pack in the next iteration; Step 4-5: Determine whether the optimization conditions are met. If the number of iterations has not reached T, and the fitness value fit error of two consecutive trainings is not less than the maximum expected error value ε, return to step 4-2; otherwise, output the optimal gray wolf position vector X α , execute steps 4-6; Step 4-6: Output the optimal gray wolf position vector X according to step 4-5 α , and obtain the corresponding weights in the extreme learning machine fire warning model and threshold h, and construct an optimized fire classification warning model.

3. The mine fire graded early warning method based on intelligent swarm algorithm according to claim 1 is characterized by: The calculation formula of the evaluation model in step 6 is as follows: In the formula, TP is a true positive example, which is a positive sample predicted as positive by the model; FP is a false positive example, which is a negative sample predicted as positive by the model; FN is a false negative example, which is a positive sample predicted as negative by the model; TN is a true negative example, which is a negative sample predicted as negative by the model.

4. The mine fire graded early warning method based on intelligent swarm algorithm according to claim 1 is characterized by: The formula for normalizing all indicator data in step 2 is as follows: In the formula, E i ' is the normalized data, E i is the original data, E max and E min are the maximum and minimum values ​​in the sample data set, respectively.

5. The mine fire graded early warning method based on intelligent swarm algorithm according to claim 2 is characterized by: The value range of g in step 4-2 is [0, 1].

6. The mine fire graded early warning method based on intelligent swarm algorithm according to claim 2 is characterized by: The value range of r1 and r2 in step 4-4 is [0, 1].

7. The mine fire graded early warning method based on intelligent swarm algorithm according to claim 1 is characterized by: The ratio of the training set to the test set in step 2 is 7:3.

Images