Mine underground precise fire prevention and extinguishing coordination system and method based on rock mass fracture identification

Abstract

The application discloses a mine underground precise fire prevention and extinguishing coordination system based on rock mass fracture identification, comprising a data acquisition module, a rock mass fracture identification module and a risk assessment module, and discloses a mine underground precise fire prevention and extinguishing coordination method based on rock mass fracture identification.The fire prevention and extinguishing coordination system and method designed by the application realize the whole process from image input to fracture feature extraction, fracture positioning and risk assessment, can effectively analyze the connection between fire risk and fracture performance characteristics, improve the identification accuracy of coal rock fractures and the reliability of risk assessment, and design an effective fire prevention and extinguishing scheme in combination with expert experience.In addition, a decision tree model is established, fire risk is scored according to image fracture characteristics, and finally, an underground coal fire comprehensive prevention and control theory and technology system is constructed, so that precise targeted treatment of underground coal fire can be realized.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of coal mine fire control, and particularly provides a mine underground precise fire prevention and extinguishing collaborative system and method based on rock mass fracture identification. BACKGROUND

[0002] The coal spontaneous combustion in the underground coal mine is mainly that the oxygen in the fresh air flow spreads to the goaf through the gap of the coal rock, the residual coal in the goaf encounters the oxygen to cause oxidation, heating, temperature rise and even combustion, and the underground coal fire is formed by the continuous development.

[0003] The existing underground fire prevention and extinguishing mainly adopts grouting technology, nitrogen injection technology, inert gas technology, group agent technology and goaf sealing technology. Although the existing underground fire prevention and extinguishing technology of the coal mine improves the safety of the coal mine to a certain extent, the precise identification of the underground rock mass fracture is not considered generally, which mainly shows that:

[0004] (1) the existing technology mostly depends on experience judgment or macroscopic identification, and cannot effectively locate the small or hidden fractures, so that the fire extinguishing measures are insufficient;

[0005] (2) the fracture identification is not accurate, and the fire prevention and extinguishing measures such as grouting and gas injection may cause waste of resources or uneven distribution, affecting the overall fire prevention and extinguishing effect;

[0006] (3) the fracture characteristics are not understood in depth, which may cause the protection measures to be unable to completely block the fractures around the fire source, and the fire risk cannot be completely eliminated.

[0007] Based on the above problems, the characteristics of the underground rock mass fracture of the mine are accurately identified, the danger of the fracture is evaluated in advance, and the targeted fire prevention and extinguishing scheme is designed according to the actual distribution of the fracture, so as to effectively make up for the shortcomings of the existing technology, realize the precision of the fire prevention and extinguishing measures and the optimal allocation of resources. SUMMARY

[0008] To solve the above problems, the present application provides a mine underground precise fire prevention and extinguishing collaborative system and method based on rock mass fracture identification, which can analyze the fracture development of the underground engineering rock mass through the machine learning algorithm, predict and judge the fracture development characteristics inside the rock, construct a representation model of the underground coal fire prevention and extinguishing technology according to the action mechanism of the underground coal fire disaster prevention, construct a comprehensive prevention and control technology system of the underground coal fire according to the characteristics of each stage of the underground coal fire development under different conditions, and realize the precise targeted treatment of the underground coal fire.

[0009] To achieve the above purpose, the technical scheme adopted by the present application is that the mine underground precise fire prevention and extinguishing collaborative system based on rock mass fracture identification comprises a data acquisition module, a rock mass fracture identification module and a risk assessment module.

[0010] The data acquisition module is used for acquiring coal rock fracture development related influencing factors and underground coal rock image data;

[0011] The rock mass fracture identification module is used for constructing a D-Net model network to accurately locate specific positions of fractures in images and calculating the number, length, width, depth, maximum length of longitudinal extension and maximum length of transverse extension of the fractures.

[0012] The risk assessment module performs risk assessment on the fractures according to the results obtained by the rock mass fracture identification module and fire risk scoring of the industry experts based on the image fracture characteristics.

[0013] Further, the coal rock fracture development related influencing factors include geological stress conditions, physical and mechanical properties of the coal rock itself, fluid action, temperature change, engineering disturbance, time effect and coal rock chemical effect.

[0014] Further, the rock mass fracture identification module includes an Encoder module, a Decoder module, a DropBlock module, a DRB module and a feature fusion module.

[0015] The Encoder module represents scale features through pooling.

[0016] The Decoder module represents scale features through up-sampling.

[0017] The DropBlock module is used for processing convolution features and inhibiting overfitting of the convolutional neural network.

[0018] The DRB module selects the best path in multiple paths according to the dynamic distribution of input features through a learnable weight or routing function while maintaining the global balance of the network.

[0019] The feature fusion module is used for improving the expression ability of the model.

[0020] The mine underground precise fire prevention and extinguishing collaboration method based on rock mass fracture identification adopts the mine underground precise fire prevention and extinguishing collaboration system and method based on rock mass fracture identification, and specifically includes the following steps:

[0021] Step one, acquiring coal rock fracture development related influencing factors and underground coal rock images through the data acquisition module.

[0022] Step two, cutting the underground coal rock images to ensure the same size, enhancing the image binarization and adaptive histogram equalization to improve the fracture clarity, and further improving the sensitivity and recognition accuracy of the rock mass fracture identification module to the fracture characteristics.

[0023] Step three, use image annotation software to label the cracks, and divide the labeled data set into training set and test set;

[0024] Step four, build a D-Net model network, which contains multiple convolutional layers and pooling layers, for extracting features from images of different scales;

[0025] Step five, the D-Net model network contains an Encoder module and a Decoder module, each module has four scale feature extraction structures, the Encoder module extracts scale features through pooling, and the Decoder module extracts scale features through upsampling;

[0026] Step six, each feature extraction structure model contains convolutional layers and DropBlock modules, and introduces DRB modules, which make full use of limited labeled data to train deeper networks and extract more complex features;

[0027] Step seven, after each level of feature extraction, the multi-scale features are integrated through a feature fusion module to form a more representative feature representation, combining low-level detail features with high-level semantic features to enhance the feature representation ability;

[0028] Step eight, after feature fusion, use convolutional layers for image segmentation and calculate loss for backpropagation to accurately locate the specific position of the crack in the image;

[0029] Step nine, after training the D-Net model network on the training set, the model weight parameters are obtained, which will be used to segment the cracks in the image during model testing;

[0030] Step ten, calculate the number, length, width, depth, maximum length of longitudinal extension and maximum length of transverse extension of the crack;

[0031] Step eleven, introduce industry experts to score the fire risk according to the image crack features, and divide it into first-level risk, second-level risk, third-level risk and fourth-level risk according to the risk level, with four risk levels corresponding to extremely high risk, higher risk, intermediate risk and lower risk, respectively. Combined with professional knowledge and experience, provide senior judgment on crack risk;

[0032] Step twelve, take the number, length, width, depth, maximum length of longitudinal extension and maximum length of transverse extension of the crack as input, and the expert score as output, and establish a decision tree model in the risk assessment module;

[0033] Step three, for other downhole coal and rock image data, the D-Net model network can be used to segment out the fissures and calculate the fissure index, and the fissure index is input into the decision tree model of the risk assessment module to obtain the fissure risk evaluation result, and then the precise targeted treatment of the underground coal fire is carried out.

[0034] Further, in step six, the complete formula of the DropBlock module is represented as:

[0035] , ;

[0036] Wherein:

[0037] is the element of the output feature map;

[0038] is the element of the input feature map;

[0039] is the mask matrix generated by random and block operation;

[0040] is the mask matrix;

[0041] represents the total probability of being zero in the entire feature map;

[0042] In order to better control the regularization strength, the DropBlock module usually dynamically adjusts the drop probability During the training process, It will gradually increase from a lower value to a target value The formula is:

[0043] ;

[0044] Wherein:

[0045] is the maximum drop probability;

[0046] is the current training step;

[0047] is the total training step;

[0048] γ is a hyperparameter that controls the drop growth rate.

[0049] Further, in step six, the formula of the output of the DRB module is represented as:

[0050] ;

[0051] Meanwhile, the optimization objective contains a loss function and a regularization term:

[0052] ;

[0053] wherein:

[0054] Y is the output of the DRB module;

[0055] z i is the activation value of path i;

[0056] z j is the activation value of path j;

[0057] P i (X) is the processing result of path i on input feature x;

[0058] is the objective function term of optimization;

[0059] is the loss of the main task;

[0060] is a balance hyperparameter used to adjust the regularization strength.

[0061] Further, in step seven, the general formula of the feature fusion module is represented as:

[0062] ;

[0063] wherein:

[0064] is the fusion of multiple features into an output feature;

[0065] is the i-th input feature map;

[0066] n is the n-th element;

[0067] is the dynamically generated weight;

[0068] represents the feature aggregation method.

[0069] Further, in step eight, the backpropagation is used to calculate the gradient of the loss function on the parameters layer by layer from the output layer, and then the weights are updated using the optimization algorithm. The core formula includes the chain rule for calculating the error term, the weight gradient, and the bias gradient. This mechanism efficiently realizes the training of neural networks and is the basis for optimizing deep learning models.

[0070] The recursive formula of backpropagation is:

[0071] a, error term

[0072] ;

[0073] in:

[0074] For the first Layer error terms;

[0075] L represents the last layer;

[0076] The activation value of layer L;

[0077] The derivative of the activation function;

[0078] This is the weight matrix;

[0079] For the first Layer error terms;

[0080] b. Weight gradient

[0081] ;

[0082] in:

[0083] The weight gradient of the l-th layer;

[0084] This is the weight matrix;

[0085] The activation value of the previous layer;

[0086] c. Bias gradient

[0087] ;

[0088] in:

[0089] For the first The bias gradient of the layer;

[0090] It is the bias vector;

[0091] For the first Error terms of the layer.

[0092] Furthermore, in step thirteen, the precise targeted treatment of the underground coal fire is as follows:

[0093] Level 1 risks are addressed through a comprehensive approach to prevention and control in the risk area, utilizing water, inhibitors, slurry, liquid inert gas, gel, three-phase foam, and ultrafine water mist.

[0094] The secondary risk adopts water body, slurry, inert gas, liquid inert gas, physical inhibitor, three-phase foam and ultra-fine water mist to prevent and treat the risk area.

[0095] The tertiary risk adopts three-phase foam, slurry, coal oxidation inhibitor, solidified foam and colloid to prevent and treat the risk area.

[0096] The fourth risk adopts inert gas, liquid inert gas, three-phase foam and ultra-fine water mist to prevent and treat the risk area.

[0097] The beneficial effects of using the present application are:

[0098] The fire prevention and extinguishing collaborative system and method designed in the present application realize the whole process from image input to crack feature extraction, crack positioning and risk assessment, can effectively analyze the connection between fire risk and crack performance characteristics, improve the identification accuracy of coal rock cracks and the reliability of risk assessment, and design effective fire prevention and extinguishing schemes combined with expert experience.

[0099] In addition, a decision tree model is established, fire risk is scored according to image crack features, and the optimal combination of disaster prevention medium and the collaborative implementation scheme suitable for the characteristics of each stage of underground coal fire development under different conditions are researched, so as to research the precise optimization method of disaster prevention medium based on the combined inhibition of multi-physical field coupling process, construct the comprehensive prevention and control theory and technology system of underground coal fire, and realize the precise targeted treatment of underground coal fire. BRIEF DESCRIPTION OF DRAWINGS

[0100] Figure 1 The flow chart of the present application is based on the precise fire prevention and extinguishing collaborative method of underground mine.

[0101] Figure 2 The schematic diagram of the present application is based on the precise fire prevention and extinguishing collaborative system of underground mine.

[0102] Figure 3 The D-Net model network diagram of the present application.

[0103] Figure 4 The schematic diagram of the present application is based on the precise targeted treatment of underground coal fire. DETAILED DESCRIPTION

[0104] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0105] Embodiment one

[0106] With reference to Figure 2 , the underground mine precision fire prevention and extinguishing coordination system based on rock mass fracture identification comprises a data acquisition module, a rock mass fracture identification module and a risk assessment module.

[0107] The data acquisition module is used to acquire coal rock fracture development related influencing factors and underground coal rock image data.

[0108] The rock mass fracture identification module is used to construct a D-Net model network to accurately locate the specific position of the fracture in the image, and to calculate the number, length, width, depth, maximum length of longitudinal extension and maximum length of transverse extension of the fracture.

[0109] The risk assessment module assesses the risk of the fracture according to the results obtained by the rock mass fracture identification module and the fire risk scoring of the image fracture characteristics by industry experts.

[0110] For underground coal mines, coal spontaneous combustion must have sufficient oxygen, and fractures are the channels for oxygen. The oxygen in fresh air flow interacts with broken coal through these gaps, and the coal body oxidizes and heats up until it burns, causing coal seam spontaneous combustion, which continues to develop into underground coal fires. Therefore, fracture identification is very important in underground fire prevention and extinguishing work.

[0111] Through the data acquisition module, fracture-related data and coal rock images are collected. Through the rock mass fracture identification module, the specific position of the fracture is accurately located, and the number, length, width, depth, maximum length of longitudinal extension and maximum length of transverse extension of the fracture are calculated. Finally, through the analysis and scoring of the risk assessment module and industry experts, according to the characteristics of each stage of underground coal fire development under different conditions, the optimal combination and coordinated implementation scheme of the disaster prevention medium suitable for it are studied, so as to study the precision optimization method of the disaster prevention medium based on the combined inhibition of the multi-physical field coupling process, construct the comprehensive prevention and control theory and technology system of underground coal fire, and realize the precise targeted treatment of underground coal fire.

[0112] Specifically, the coal rock fracture development related influencing factors include geological stress conditions, physical and mechanical properties of coal rock itself, fluid action, temperature change, engineering disturbance, time effect and coal rock chemical effect.

[0113] The geological stress conditions include high stress and differential stress, the physical and mechanical properties of coal rock itself include the strength and elastic modulus of coal rock, the initial fracture distribution and the bedding and structural plane of rock, the fluid action includes the groundwater pressure, coalbed methane pressure and fluid-induced chemical reaction, and the engineering disturbance includes blasting effect, surrounding rock rupture and collapse and stress redistribution caused by mining.

[0114] High ground stress, deep coal and rock are controlled by three-dimensional ground stress, among which the maximum principal stress (σ1) plays a leading role in the initial formation and expansion of fractures. High ground stress leads to significant compaction of coal and rock, and part of the fractures are closed. However, when the stress exceeds the strength limit of the rock mass, shear fractures or tensile fractures will be triggered.

[0115] Stress difference, the greater the stress difference, the more likely the rock mass will produce significant shear fractures, accompanied by the development of local tensile fractures.

[0116] Strength and elastic modulus of coal and rock, the lower the strength and the smaller the elastic modulus, the easier it is for coal and rock to form fractures and expand.

[0117] Initial fracture distribution, coal and rock naturally exist fracture network, the distribution, density and scale of initial fracture determine the path and rule of fracture expansion.

[0118] Rock bedding and structural plane, weak planes (such as bedding, joints) in coal and rock are easy to become the starting point of fracture development.

[0119] Groundwater pressure, fluid pressure (pore water pressure) in the fracture will produce normal support force on the fracture wall, reducing the effective stress of the rock mass, thus promoting the expansion of the fracture.

[0120] Coalbed methane pressure, the presence and flow of methane and other gases in coal seams cause additional gas pressure on the fracture, increasing the tendency of the fracture to open.

[0121] Fluid-induced chemical reaction, the fracture of coal and rock may be eroded or dissolved under the action of water chemistry, further expanding the fracture width.

[0122] Temperature change, fracture expansion is intensified in high temperature environment, especially in the process of coal spontaneous combustion, a large number of fractures will be generated, providing channels for oxygen transport and coal combustion.

[0123] Blasting effect, blasting vibration can form blasting fractures, and cause the expansion and connection of existing fractures.

[0124] Stress redistribution caused by mining, mining disturbance leads to increased fracture density, and fractures extend along the direction of maximum stress.

[0125] Surrounding rock failure and collapse, engineering disturbance causes local instability of coal and rock, and significant development of fractures.

[0126] Time effect, long-term creep may cause gradual expansion of fractures, eventually connecting, and reducing the overall stability of the rock mass.

[0127] Coal and rock chemical reaction, oxidation may weaken the strength of coal, and induce fracture development.

[0128] Specifically, the rock mass fracture identification module includes an Encoder module, a Decoder module, a DropBlock module, a DRB module, and a feature fusion module.

[0129] The Encoder module represents scale features through pooling, for the encoder.

[0130] The Decoder module represents scale features through up-sampling, for the decoder.

[0131] The DropBlock module is used to process convolutional features and suppress overfitting of the convolutional neural network.

[0132] The DRB module selects the best path among multiple paths according to the dynamic distribution of input features through a learnable weight or routing function, while maintaining the global balance of the network.

[0133] The feature fusion module is used to improve the expression ability of the model.

[0134] Embodiment Two

[0135] As shown in Figure 1 , the mine underground precision fire prevention and extinguishing collaboration method based on rock mass fracture identification adopts the mine underground precision fire prevention and extinguishing collaboration system and method based on rock mass fracture identification in Embodiment One, and specifically includes the following steps:

[0136] Step One: Collect coal and rock fracture development related influencing factors and underground coal and rock images through the data acquisition module.

[0137] The images come from underground monitoring and coal and rock surface field photos, ensuring that the images contain fracture information of different scales of coal pillars and roofs.

[0138] Step Two: Cut the underground coal and rock images to ensure the same size, and enhance the image binarization and adaptive histogram equalization to improve the fracture clarity, and further improve the sensitivity and recognition accuracy of the rock mass fracture identification module to fracture features.

[0139] Step Three: Use image annotation software to annotate the fractures, and divide the annotated data set into a training set and a test set.

[0140] The training set is used to train the model, and the test set is used to evaluate the pros and cons of the model.

[0141] Step Four: As shown in Figure 3 , a D-Net model network is constructed, which includes multiple convolutional layers and pooling layers, and is used to extract features from images of different scales.

[0142] Step five, the D-Net model network includes an Encoder module and a Decoder module, each module has a feature extraction structure of 4 scales, the Encoder module represents scale features through pooling, and the Decoder module represents scale features through upsampling;

[0143] Step six, the model of each feature extraction structure contains a convolutional layer and a DropBlock module, and a DRB module is introduced, which makes full use of limited labeled data to train deeper networks and extract more complex features;

[0144] DropBlock is a regularization technique in deep learning, which is used to suppress the overfitting of convolutional neural networks, especially by randomly discarding blocks of regions on the convolutional feature map. Unlike traditional Dropout, DropBlock is more suitable for processing convolutional features because it continuously masks a certain region, forcing the network to pay attention to more extensive contextual information.

[0145] The complete formula of the DropBlock module is:

[0146] , ;

[0147] Where:

[0148] is the element of the output feature map;

[0149] is the element of the input feature map;

[0150] is the mask matrix after random generation and block operation;

[0151] is the mask matrix;

[0152] represents the total probability of being zero in the entire feature map;

[0153] By giving the input feature map and the parameters of DropBlock, the output feature map is obtained, where n, c, h, w represent: n samples, c channels, and spatial size h x w;

[0154] In order to better control the regularization strength, the DropBlock module usually dynamically adjusts the drop probability , which gradually increases from a lower value to a target value during training , the formula is:

[0155] ;

[0156] wherein:

[0157] is the maximum dropping probability;

[0158] is the current training step;

[0159] is the total training step;

[0160] γ is a hyper-parameter to control the growth rate of dropping;

[0161] The core idea of the DRB module is to select the best path among multiple paths (or subnetworks) according to the dynamic distribution of input features through a learnable weight or routing function, while maintaining the global balance of the network;

[0162] The formula representation of the output of the DRB module is:

[0163] ;

[0164] At the same time, the optimization objective contains the loss function and the regularization term:

[0165] ;

[0166] wherein:

[0167] Y is the output of the DRB module;

[0168] z i is the activation value of path i;

[0169] z j is the activation value of path j;

[0170] P i (X) is the processing result of path i on input feature x;

[0171] is the objective function term of optimization;

[0172] is the loss of the main task;

[0173] is the balance hyper-parameter, used to adjust the regularization strength;

[0174] Step seven, after feature extraction at each level, the multi-scale features are integrated through a feature fusion module to form more representative feature representations, combining low-level detailed features with high-level semantic features to enhance the representation ability of the features;

[0175] The feature fusion module is a core component in deep learning used to combine feature maps from different sources, scales, or network paths. Feature fusion can usually improve the expressive power of the model and is especially important in tasks such as image classification, object detection, and semantic segmentation.

[0176] The general formula for the feature fusion module is expressed as:

[0177] ;

[0178] in:

[0179] Multiple features are fused into a single output feature;

[0180] Let i be the i-th input feature map;

[0181] n is the nth element;

[0182] The weights are dynamically generated.

[0183] Indicate the feature aggregation method, use directly. Or it may undergo transformations such as upsampling, downsampling, or convolution;

[0184] Step 8: After feature fusion, use convolutional layers to segment the image and calculate the loss for backpropagation to accurately locate the specific position of the crack in the image.

[0185] Backpropagation is a core algorithm in deep learning for optimizing the weights of neural networks. By calculating the gradient of the loss function with respect to each parameter, backpropagation guides parameter updates, ultimately achieving optimal model training. The following is a detailed derivation and analysis of the backpropagation formula;

[0186] By using a recursive formula, the gradient of the loss function with respect to the parameters is calculated layer by layer starting from the output layer. Then, the weights are updated using an optimization algorithm. Its core formulas include the chain rule for calculating the error term, weight gradient, and bias gradient. This mechanism efficiently realizes the training of neural networks and is the foundation for the optimization of deep learning models.

[0187] The recursive formula for backpropagation is:

[0188] a. Error Term

[0189] ;

[0190] in:

[0191] For the first Layer error terms;

[0192] L represents the last layer;

[0193] The activation value of layer L;

[0194] The derivative of the activation function;

[0195] This is the weight matrix;

[0196] For the first Layer error terms;

[0197] b. Weight gradient

[0198] ;

[0199] in:

[0200] The weight gradient of the l-th layer;

[0201] This is the weight matrix;

[0202] The activation value of the previous layer;

[0203] T is a mathematical symbol that means transpose;

[0204] c. Bias gradient

[0205] ;

[0206] in:

[0207] For the first The bias gradient of the layer;

[0208] It is the bias vector;

[0209] For the first Layer error terms;

[0210] Step 9: After training the D-Net model on the training set, the model weight parameters are obtained. In the model test, the weights will be used to segment the cracks in the image.

[0211] Step 10: Calculate the number, length, width, depth, maximum longitudinal extension length, and maximum transverse extension length of the cracks;

[0212] The previous steps aimed to identify cracks, but the shape and number of cracks are different in each image. Therefore, it is necessary to calculate indicators about cracks, such as the number, length, width, depth, maximum longitudinal extension length, and maximum lateral extension length of cracks.

[0213] Step eleven, introduce industry experts to score fire risk according to image crack features, divide into first-level risk, second-level risk, third-level risk, fourth-level risk according to risk degree, four risk levels correspond to extremely high risk, higher risk, intermediate risk, lower risk in turn, combined with professional knowledge and experience, provide senior judgment on crack risk;

[0214] Step twelve, taking the number, length, width, depth, maximum length of longitudinal extension and maximum length of transverse extension of the crack as input, and expert score as output, a decision tree model is established in the risk assessment module;

[0215] Step thirteen, for other underground coal rock image data, the D-Net model network can be used to segment the cracks and calculate the crack indicators, and the crack indicators can be input into the decision tree model of the risk assessment module to obtain the crack risk evaluation result;

[0216] Through the above steps, the whole process from image input to crack feature extraction, crack positioning and risk assessment is realized, and the recognition accuracy of coal rock cracks and the reliability of risk assessment are improved;

[0217] According to the image crack feature, the fire risk is scored, and according to the characteristics of each stage of underground coal fire development under different conditions, the optimal combination and collaborative implementation scheme of disaster prevention medium suitable for the characteristics are studied, so as to study the precise optimization method of disaster prevention medium based on the combined inhibition of multi-physical field coupling process, construct the comprehensive prevention and control theory and technology system of underground coal fire, and realize the precise targeted treatment of underground coal fire;

[0218] As shown in Figure 4 , the precise targeted treatment of underground coal fire is:

[0219] The first-level risk adopts water body, inhibitor, slurry, liquid inert gas, gel, three-phase foam and ultra-fine water mist for comprehensive prevention and control of the risk area;

[0220] The second-level risk adopts water body, slurry, inert gas, liquid inert gas, physical inhibitor, three-phase foam and ultra-fine water mist for prevention and control of the risk area;

[0221] The third-level risk adopts three-phase foam, slurry, coal oxidation inhibitor, solidified foam and colloid for prevention and control of the risk area;

[0222] The fourth-level risk adopts inert gas, liquid inert gas, three-phase foam and ultra-fine water mist for prevention and control of the risk area.

[0223] The above merely describes the preferred embodiments of the present application, and for those skilled in the art, many changes can be made to the specific implementation and application range according to the idea of the present application, as long as these changes do not deviate from the concept of the present application, and all of them belong to the protection scope of the present application.

Claims

1. A method for precise fire prevention and extinguishing in a mine based on rock mass fracture identification, comprising a precise fire prevention and extinguishing system in a mine based on rock mass fracture identification, the system comprising a data acquisition module, a rock mass fracture identification module, and a risk assessment module; the data acquisition module is used to acquire coal rock fracture development related influencing factors and underground coal rock image data; the rock mass fracture identification module is used to construct a D-Net model network to accurately locate the specific position of the fracture in the image, and calculate the number, length, width, depth, maximum length of longitudinal extension and maximum length of transverse extension of the fracture; the risk assessment module assesses the risk of the fracture according to the results obtained by the rock mass fracture identification module and the fire risk score of the image fracture characteristics by industry experts; the rock mass fracture identification module comprises an Encoder module, a Decoder module, a DropBlock module, a DRB module, and a feature fusion module; the Encoder module expresses the scale feature through pooling; the Decoder module expresses the scale feature through upsampling; the DropBlock module is used to process the convolution feature and suppress the overfitting of the convolutional neural network; the DRB module selects the best path in multiple paths according to the dynamic distribution of the input feature through a learnable weight or routing function, while maintaining the global balance of the network; the feature fusion module is used to improve the expression ability of the model; Specifically comprising the following steps: Step one, collect coal rock fracture development related influencing factors and underground coal rock images through the data acquisition module; Step two, shear the underground coal rock image to ensure the same size, and enhance the fracture clarity through image binarization and adaptive histogram equalization, thereby improving the sensitivity and recognition accuracy of the rock mass fracture identification module to the fracture characteristics; Step three, use image annotation software to label the fracture, and divide the labeled data set into a training set and a test set; Step four, construct a D-Net model network containing multiple convolution layers and pooling layers for extracting features from images of different scales; Step five, the D-Net model network contains an Encoder module and a Decoder module, each module has four scale feature extraction structures, the Encoder module expresses the scale feature through pooling, and the Decoder module expresses the scale feature through upsampling; Step six, each feature extraction structure model contains a convolution layer and a DropBlock module, and a DRB module is introduced, which makes full use of limited labeled data to train deeper networks and extract more complex features; Step seven, after each level of feature extraction, the multi-scale features are integrated through the feature fusion module to form more representative feature representations, combining low-level detail features with high-level semantic features to enhance the feature representation ability; Step eight, after feature fusion, use the convolution layer for image segmentation and calculate the loss for back propagation to accurately locate the specific position of the fracture in the image. Step nine, after the D-Net model network is trained on the training set, the model weight parameters are obtained, and the weight will be used in the model test to segment the cracks in the image; Step ten, the number, length, width, depth, maximum length of longitudinal extension and maximum length of transverse extension of the cracks are calculated; Step eleven, industry experts introduce fire risk scoring according to the image crack characteristics, and the risk is divided into first-level risk, second-level risk, third-level risk and fourth-level risk according to the risk level, and the four risk levels correspond to extremely high risk, higher risk, intermediate risk and lower risk in turn, combined with professional knowledge and experience, a high-level judgment of crack risk is provided; Step twelve, the number, length, width, depth, maximum length of longitudinal extension and maximum length of transverse extension of the cracks are taken as inputs, and the expert scoring is taken as output, and a decision tree model is established in the risk assessment module; Step thirteen, for other underground coal rock image data, the cracks can be segmented by the D-Net model network and the crack indicators are calculated, the crack indicators are input into the decision tree model of the risk assessment module, and the crack risk evaluation result is obtained, and then the precise targeted treatment of underground coal fire is carried out.

2. The mine underground precision fire prevention and extinguishing collaborative method based on rock mass fracture identification according to claim 1, characterized in that: The coal rock crack development related influencing factors include geological stress conditions, physical and mechanical properties of coal rock itself, fluid action, temperature change, engineering disturbance, time effect and coal rock chemical effect.

3. The rock mass fracture identification-based precise fire prevention and extinguishing coordination method in a mine shaft according to claim 1, characterized in that: In step six, the complete formula of the DropBlock module is represented as: ， ； Wherein: is an element of the output feature map; is an element of the input feature map; is a mask matrix that has been randomly generated and block operated; is a mask matrix; denotes the total probability of being zeroed in the whole feature map; DropBlock module dynamically adjusts the drop probability During training, from a lower value to the target value The formula is: ； Wherein: is the maximum drop probability; current_train_steps is the current training step number; is the total number of training steps; γ is a hyperparameter for controlling the drop growth rate.

4. The rock mass fracture identification-based precise fire prevention and extinguishing coordination method in a mine shaft according to claim 1, characterized in that: In step six, the formula of the output of the DRB module is represented as: ； At the same time, the optimization target includes a loss function and a regularization term: ； Wherein: Y is the output of the DRB module; z i is the activation value for path i; z j is the activation value for path j; P i (X) is the result of processing of input feature x by path i; for an optimized objective function term; loss for the main task; To balance the hyperparameters, used to adjust the regularization strength.

5. The rock mass fracture identification-based precise fire prevention and extinguishing coordination method in a mine shaft according to claim 1, characterized in that: In step seven, the general formula of the feature fusion module is represented as: ； Wherein: to fuse a plurality of features into one output feature; is the i-th input feature map; n is the nth element; are dynamically generated weights; represents a characteristic aggregation mode.

6. The rock mass fracture identification-based precise fire prevention and extinguishing coordination method in a mine shaft according to claim 1, characterized in that: In step eight, the back propagation is calculated layer by layer from the output layer to the loss function of the parameter by using the recursive formula, and then the weight is updated by using the optimization algorithm, and the core formula includes the chain rule calculation error term, weight gradient and bias gradient; The recursive formula of back propagation is: a, error term ； Wherein: for the first error term for the layer; L is the last layer; L is the activation value for the L layer; derivative of the activation function; is a weight matrix; for the first layer error term; b, weight gradient ； Wherein: For the first weight gradient of the layer; is a weight matrix; is the activation value of the previous layer; c, bias gradient ； Wherein: For the first Bias gradient of the layer; is a bias vector; For the first layer the error term is.

7. The rock mass fracture identification-based precise fire prevention and extinguishing coordination method in a mine shaft according to claim 1, characterized in that: In step thirteen, the precise targeted treatment of underground coal fire is: The first-level risk adopts water body, inhibitor, slurry, liquid inert gas, gel, three-phase foam and ultra-fine water mist for comprehensive prevention and control of the risk area; The second-level risk adopts water body, slurry, inert gas, liquid inert gas, physical inhibitor, three-phase foam and ultra-fine water mist for prevention and control of the risk area; The third-level risk adopts three-phase foam, slurry, coal oxidation inhibitor, solidified foam and colloid for prevention and control of the risk area; The fourth-level risk adopts inert gas, liquid inert gas, three-phase foam and ultra-fine water mist for prevention and control of the risk area.

Images