Mine underground accurate fire preventing and extinguishing cooperation system and method based on rock mass fracture recognition

Abstract

The invention discloses a mine underground precise fire preventing and extinguishing cooperation system based on rock mass fracture recognition, the mine underground precise fire preventing and extinguishing cooperation system comprises a data acquisition module, a rock mass fracture recognition module and a risk assessment module, and the invention further discloses a mine underground precise fire preventing and extinguishing cooperation method based on rock mass fracture recognition. The whole process from image input to fracture feature extraction, fracture positioning and risk assessment is achieved, the relation between the fire risk and fracture performance features can be effectively analyzed, the recognition accuracy of coal rock fractures and the reliability of risk assessment are improved, an effective fire prevention and extinguishing scheme is designed in combination with expert experience, in addition, a decision-making tree model is established, and the risk assessment efficiency is improved. Fire risk scoring is carried out according to image crack characteristics, an underground coal fire comprehensive prevention and control theory and technical system is finally constructed, and accurate targeted treatment of underground coal fire can be realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of coal mine fire control, and in particular provides a precise underground mine fire prevention and extinguishing coordinated system and method based on rock mass fissure identification. Background Art

[0002] Spontaneous combustion of coal in coal mines is mainly caused by the oxygen in the fresh air flow being transmitted to the goaf through the gaps in the coal rock. The remaining coal in the goaf is oxidized by oxygen, generates heat and rises in temperature until it burns, which continues to develop into an underground coal fire.

[0003] Existing underground fire prevention and extinguishing technologies mostly use grouting technology, nitrogen injection technology, inert gas technology, chemical agent technology, and goaf sealing technology. Although the existing underground coal mine fire prevention and extinguishing technologies have improved the safety of coal mines to a certain extent, they generally do not take into account the accurate identification of underground rock mass fissures, which are mainly manifested in:

[0004] (1) Most existing technologies rely on empirical judgment or macroscopic identification and cannot effectively locate small or hidden cracks, resulting in insufficient coverage of fire extinguishing measures;

[0005] (2) Inaccurate crack identification may lead to waste or uneven distribution of resources, thus affecting the overall fire prevention and extinguishing effect.

[0006] (3) Lack of in-depth understanding of crack characteristics may result in protective measures being unable to completely seal cracks around fire sources, resulting in hidden dangers and the inability to completely eliminate fire risks.

[0007] Based on the above problems, by accurately identifying the characteristics of rock cracks in underground mines, assessing the danger of cracks in advance, and designing targeted fire prevention and extinguishing plans based on the actual distribution of cracks, we can effectively make up for the shortcomings of existing technologies and achieve precise fire prevention measures and optimal allocation of resources. Summary of the invention

[0008] To solve the above problems, the present invention provides a precise underground mine fire prevention and extinguishing coordinated system and method based on rock mass fracture identification. The system and method can analyze the development of rock mass fractures in underground engineering through machine learning algorithms, predict and judge the development characteristics of fractures inside the rock, and construct a characterization model of underground coal fire prevention and extinguishing technology according to the mechanism of action of underground coal fire disaster prevention. According to the characteristics of each stage of underground coal fire development under different conditions, a comprehensive underground coal fire prevention and control technology system is constructed to achieve precise targeted treatment of underground coal fires.

[0009] To achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a precise underground mine fire prevention and extinguishing coordination system based on rock mass fissure identification, including a data acquisition module, a rock mass fissure identification module and a risk assessment module;

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

[0011] The rock mass crack identification module is used to construct a D-Net model network to accurately locate the specific position of the crack in the image, and calculate the number, length, width, depth, maximum longitudinal extension length and maximum lateral extension length of the crack;

[0012] The risk assessment module performs risk assessment on the cracks based on the results obtained by the rock crack identification module and the fire risk scores of industry experts based on the crack characteristics of the image.

[0013] Furthermore, the influencing factors related to the development of coal rock fractures 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] Furthermore, 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 expresses scale features through pooling;

[0016] The Decoder module expresses scale features through upsampling;

[0017] The DropBlock module is used to process convolution features and suppress overfitting of convolutional neural networks;

[0018] The DRB module selects the best path among multiple paths through learnable weights or routing functions according to the dynamic distribution of input features, while maintaining the global balance of the network;

[0019] The feature fusion module is used to improve the expressiveness of the model.

[0020] The method for precise underground fire prevention and extinguishing in a mine based on rock mass crack identification adopts the above-mentioned system and method for precise underground fire prevention and extinguishing in a mine based on rock mass crack identification, and specifically includes the following steps:

[0021] Step 1: collecting influencing factors related to coal-rock fracture development and underground coal-rock images through a data acquisition module;

[0022] Step 2: Cut the underground coal and rock images to ensure the same size, and enhance the image binarization and adaptive histogram equalization to improve the clarity of the cracks, thereby improving the sensitivity and recognition accuracy of the rock crack recognition module to the crack characteristics;

[0023] Step 3: Use image annotation software to annotate the cracks and divide the annotated data set into a training set and a test set;

[0024] Step 4: Build a D-Net model network, which includes multiple levels of convolutional layers and pooling layers to extract features from images of different scales.

[0025] Step 5: The D-Net model network includes an Encoder module and a Decoder module. Each module has a feature extraction structure of four scales. The Encoder module expresses scale features through pooling, and the Decoder module expresses scale features through upsampling.

[0026] Step 6: Each feature extraction model contains a convolutional layer and a DropBlock module, and introduces a DRB module. This model makes full use of limited labeled data to train a deeper network and extract more complex features.

[0027] Step 7: After extracting features at each level, the multi-scale features are integrated through the feature fusion module to form a more representative feature representation, combining low-level detail features with high-level semantic features to enhance the representation ability of features;

[0028] Step 8: After feature fusion, use the convolutional layer to segment the image and calculate the loss for back propagation to accurately locate the specific location of the crack in the image.

[0029] Step 9: After training the D-Net model network on the training set, the model weight parameters are obtained. The weights will be used to segment the cracks in the image during the model test.

[0030] Step 10, calculating the number, length, width, depth, maximum longitudinal extension length and maximum transverse extension length of the cracks;

[0031] Step 11: Introduce industry experts to score fire risks based on the crack features of the image. The risk is divided into level 1 risk, level 2 risk, level 3 risk, and level 4 risk. The four risk levels correspond to extremely high risk, high risk, medium risk, and low risk, respectively. Combined with professional knowledge and experience, provide a high-level judgment on crack risks.

[0032] Step 12, using the number, length, width, depth, maximum longitudinal extension length and maximum transverse extension length of the cracks as inputs and expert scores as outputs, a decision tree model is established in the risk assessment module;

[0033] Step 13. For other underground coal-rock image data, the D-Net model network can be used to segment the cracks and calculate the crack indicators. The crack indicators are input into the decision tree model of the risk assessment module to obtain the crack risk assessment results, and then precise targeted treatment of underground coal fires can be carried out.

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

[0035] y n,c,h,w =x n,c,h,w ·M n,c,h,w ,

[0036] in:

[0037] y n,c,h,w is the element of the output feature map;

[0038] x n,c,h,w is the element of the input feature map;

[0039] M n,c,h,w is the mask matrix after random generation and block operation;

[0040] M is the mask matrix;

[0041] p block Represents the total probability of the entire feature map being set to zero;

[0042] In order to better control the regularization strength, the DropBlock module usually dynamically adjusts the drop probability p, which gradually increases from a lower value to the target value p during training. max , the formula is:

[0043]

[0044] in:

[0045] p max is the maximum drop probability;

[0046] Current step is the current training step number;

[0047] total steps is the total number of training steps;

[0048] γ is a hyperparameter that controls the rate at which dropout increases.

[0049] Further, in step six, the output of the DBR module is expressed as:

[0050]

[0051] At the same time, the optimization objective includes the loss function and the regularization term:

[0052]

[0053] in:

[0054] Y is the output of the DBR 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 for optimization;

[0059] Loss of the main mission;

[0060] To balance the hyperparameters, it is used to adjust the regularization strength.

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

[0062]

[0063] in:

[0064] F out Fusion of multiple features into one output feature;

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

[0066] n is the nth element;

[0067] α i The weights are dynamically generated;

[0068] Agg(F i ) represents the feature aggregation method.

[0069] Furthermore, in step eight, the back propagation calculates the gradient of the loss function to the parameters layer by layer starting from the output layer through a recursive formula, and then uses the optimization algorithm to update the weights. Its core formula includes the chain rule to calculate the error term, weight gradient and bias gradient. This mechanism efficiently realizes the training of neural networks and is the basis for deep learning model optimization.

[0070] The recursive formula for back propagation is:

[0071] a. Error term

[0072]

[0073] in:

[0074] δ (l) is the error term of the lth layer;

[0075] L is the last layer;

[0076] A (L) is the activation value of layer L;

[0077] g' (L) (Z (L) ) is the derivative of the activation function;

[0078] (W (l+1) ) T is the weight matrix;

[0079] δ (l+1) is the error term of the l+1th layer;

[0080] b. Weight gradient

[0081]

[0082] in:

[0083] is the weight gradient of the lth layer;

[0084] W (l) is the weight matrix;

[0085] A (l-1) is the activation value of the previous layer;

[0086] c. Bias gradient

[0087]

[0088] in:

[0089] is the bias gradient of the lth layer;

[0090] b (l) is the bias vector;

[0091] is the error term of the lth layer.

[0092] Furthermore, in step thirteen, the underground coal fire is precisely targeted and treated as follows:

[0093] Level 1 risk uses water, inhibitors, slurry, liquid inert gas, gel, three-phase foam, and ultrafine water mist for comprehensive prevention and control of risk areas;

[0094] Level 2 risk uses water, slurry, inert gas, liquid inert gas, physical inhibitors, three-phase foam, and ultrafine water mist to prevent and control the risk area;

[0095] Level 3 risk uses three-phase foam, slurry, coal oxidation inhibitor, solidified foam, and colloid to prevent and control the risk area;

[0096] For level 4 risk, inert gas, liquid inert gas, three-phase foam and ultrafine water mist are used to prevent and control risk areas.

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

[0098] The fire prevention and extinguishing collaborative system and method designed in the present invention realizes the whole process from image input to crack feature extraction, crack positioning and risk assessment. It can effectively analyze the relationship between fire risk and crack performance characteristics, improve the recognition accuracy of coal and rock cracks and the reliability of risk assessment, and design effective fire prevention and extinguishing plans in combination with expert experience.

[0099] In addition, a decision tree model is established to score the fire risk according to the image crack characteristics. According to the characteristics of each stage of underground coal fire development under different conditions, the optimal combination and coordinated implementation plan of disaster prevention media corresponding to it are studied. In this way, the precise optimization method of disaster prevention media based on the joint suppression of multi-physical field coupling process is studied, and a theoretical and technical system for the comprehensive prevention and control of underground coal fires is constructed, which can realize the precise targeted treatment of underground coal fires. BRIEF DESCRIPTION OF THE DRAWINGS

[0100] Figure 1 The present invention is a flow chart of the precise underground mine fire prevention and extinguishing coordinated method based on rock mass fracture identification.

[0101] Figure 2 It is a schematic diagram of the underground precise fire prevention and extinguishing coordinated system for mines based on rock mass fissure identification of the present invention.

[0102] Figure 3 Schematic diagram of the D-Net model network of the present invention.

[0103] Figure 4 Schematic diagram of the precise targeted treatment of underground coal fires according to the present invention. DETAILED DESCRIPTION

[0104] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0105] Embodiment 1

[0106] Reference Figure 2 The precise underground mine fire prevention and extinguishing coordinated system based on rock crack identification includes a data acquisition module, a rock crack identification module and a risk assessment module.

[0107] The data acquisition module is used to collect the influencing factors related to the development of coal and rock fractures and underground coal and rock image data.

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

[0109] The risk assessment module conducts risk assessment on cracks based on the results obtained by the rock crack identification module and the fire risk scores of industry experts based on the image crack characteristics.

[0110] For underground coal mines, there must be sufficient oxygen for coal to combust spontaneously. Cracks are channels for oxygen. Oxygen in the fresh air flow interacts with the broken coal body through these gaps. The coal body oxidizes and generates heat until it burns, causing the coal seam to spontaneously combust. It continues to develop into underground coal fires. Therefore, crack identification is very important in underground fire prevention and extinguishing work.

[0111] The data acquisition module is used to collect crack-related data and coal-rock images. The rock crack identification module is used to process the coal-rock images to accurately locate the specific positions of the cracks, and the number, length, width, depth, maximum longitudinal extension length, and maximum lateral extension length of the cracks are calculated. Finally, through the risk assessment module and analysis and scoring by industry experts, the characteristics of each stage of underground coal fire development under different conditions are studied, and the optimal combination and coordinated implementation plan of disaster prevention media corresponding to them are studied. In this way, the precise optimization method of disaster prevention media based on the joint suppression of multi-physical field coupling process is studied, and a theoretical and technical system for comprehensive prevention and control of underground coal fires is constructed to achieve precise targeted treatment of underground coal fires.

[0112] Specifically, the factors affecting the development of coal rock fractures include geological stress conditions, physical and mechanical properties of the coal rock itself, fluid action, temperature changes, engineering disturbances, time effects, and coal rock chemical effects.

[0113] Geological stress conditions include high ground stress and differential stress. The physical and mechanical properties of coal rock itself include the strength and elastic modulus of coal rock, initial crack distribution and rock stratification and structural plane. Fluid action includes groundwater pressure, coalbed methane pressure and fluid-induced chemical reactions. Engineering disturbances include blasting, surrounding rock fracture and collapse, and stress redistribution caused by mining.

[0114] High geostress. Deep coal rock is controlled by three-dimensional geostress, among which the maximum principal stress (σ1) plays a dominant role in the initial formation and expansion of cracks. High geostress causes significant compaction of coal rock and closure of some cracks. However, when the stress exceeds the strength limit of the rock mass, shear cracks or tension cracks will be triggered.

[0115] Stress difference: the greater the stress difference, the more the rock mass failure pattern tends to produce significant shear cracks, accompanied by the development of local tensile cracks.

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

[0117] Initial crack distribution. Coal rock naturally has a crack network. The distribution, density and scale of the initial cracks determine the path and law of crack expansion.

[0118] The bedding and structural planes of rocks, and weak planes in coal rocks (such as bedding and joints) can easily become the starting points for the development of cracks.

[0119] Groundwater pressure and fluid pressure in the cracks (pore water pressure) will produce normal support force on the crack wall, reduce the effective stress of the rock mass, and thus promote crack expansion.

[0120] Coalbed methane pressure, the storage and flow of gases such as methane in coal seams cause the cracks to be subjected to additional gas pressure, increasing the tendency of the cracks to open.

[0121] Fluids induce chemical reactions, and cracks in coal rocks may be eroded or dissolved under the action of water chemistry, further expanding the width of the cracks.

[0122] Temperature changes and crack expansion intensify in high temperature environments, especially during coal spontaneous combustion, when a large number of cracks will be generated, providing channels for oxygen transport and coal combustion.

[0123] The blasting effect and the impact of blasting vibration on coal rock will form blasting cracks, and at the same time cause the expansion and penetration of existing cracks.

[0124] Mining causes stress redistribution, mining disturbance leads to increased crack density, and cracks extend along the direction of maximum stress.

[0125] The surrounding rock fractures and collapses, and engineering disturbances cause local instability of the coal rock and significant development of cracks.

[0126] Time effect: Long-term creep may cause cracks to gradually expand and eventually penetrate, reducing the overall stability of the rock mass.

[0127] Coal-rock chemical reactions and oxidation may weaken the strength of the coal body and induce the development of cracks.

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

[0129] The Encoder module represents scale features through pooling and serves as an encoder.

[0130] The Decoder module represents the scale feature through upsampling and is a decoder.

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

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

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

[0134] Embodiment 2

[0135] like Figure 1 As shown, the underground precise fire prevention and extinguishing coordination method for mines based on rock mass fissure identification adopts the underground precise fire prevention and extinguishing coordination system and method for mines based on rock mass fissure identification in Example 1, and specifically includes the following steps:

[0136] Step 1: collecting influencing factors related to coal-rock fracture development and underground coal-rock images through a data acquisition module;

[0137] The images are obtained from underground monitoring and on-site photos of the coal and rock surface, ensuring that the images contain information on fractures of different scales in the coal pillars and roof;

[0138] Step 2: Cut the underground coal and rock images to ensure the same size, and enhance the image binarization and adaptive histogram equalization to improve the clarity of the cracks, thereby improving the sensitivity and recognition accuracy of the rock crack recognition module to the crack characteristics;

[0139] Step 3: Use image annotation software to annotate the cracks 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 quality of the model;

[0141] Step 4: Figure 3 As shown in the figure, a D-Net model network is constructed, which includes multi-level convolutional layers and pooling layers to extract features from images of different scales;

[0142] Step 5: The D-Net model network includes an Encoder module and a Decoder module. Each module has a feature extraction structure of four scales. The Encoder module expresses scale features through pooling, and the Decoder module expresses scale features through upsampling.

[0143] Step 6: Each feature extraction model contains a convolutional layer and a DropBlock module, and introduces a DRB module. This model makes full use of limited labeled data to train a deeper network and extract more complex features.

[0144] DropBlock is a regularization technique in deep learning that is used to suppress 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 area, forcing the network to focus on a wider range of contextual information.

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

[0146]

[0147] in:

[0148] y n,c,h,w is the element of the output feature map;

[0149] x n,c,h,w is the element of the input feature map;

[0150] M n,c,h,w is the mask matrix after random generation and block operation;

[0151] M is the mask matrix;

[0152] p block Represents the total probability of the entire feature map being set to zero;

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

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

[0155]

[0156] in:

[0157] pmax is the maximum drop probability;

[0158] Current step is the current training step number;

[0159] total steps is the total number of training steps;

[0160] γ is a hyperparameter that controls the rate of dropout growth;

[0161] The core idea of ​​the DBR module is to select the best path among multiple paths (or sub-networks) based on the dynamic distribution of input features through learnable weights or routing functions, while maintaining the global balance of the network;

[0162] The output of the DBR module is expressed as:

[0163]

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

[0165]

[0166] in:

[0167] Y is the output of the DBR 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 for optimization;

[0172] Loss of the main mission;

[0173] To balance the hyperparameters, it is used to adjust the regularization strength;

[0174] Step 7: After extracting features at each level, the multi-scale features are integrated through the feature fusion module to form a more representative feature representation, combining low-level detail features with high-level semantic features to enhance the representation ability of features;

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

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

[0177]

[0178] in:

[0179] F out Fusion of multiple features into one output feature;

[0180] F i is the i-th input feature map;

[0181] n is the nth element;

[0182] α i The weights are dynamically generated;

[0183] Agg(F i ) represents the feature aggregation method, directly using F i Or after upsampling, downsampling, convolution and other transformations;

[0184] Step 8: After feature fusion, use the convolutional layer to segment the image and calculate the loss for back propagation to accurately locate the specific location of the crack in the image.

[0185] Back propagation is the core algorithm for optimizing neural network weights in deep learning. By calculating the gradient of the loss function relative to each parameter, back propagation guides parameter updates and ultimately achieves optimal model training. The following is a detailed derivation and analysis of the back propagation formula;

[0186] Through the recursive formula, the gradient of the loss function to the parameters is calculated layer by layer starting from the output layer, and then the weight is updated using the optimization algorithm. Its core formula includes the chain rule to calculate the error term, weight gradient and bias gradient. This mechanism efficiently realizes the training of neural networks and is the basis for deep learning model optimization.

[0187] The recursive formula for back propagation is:

[0188] a. Error term

[0189]

[0190] in:

[0191] δ (l) is the error term of the lth layer;

[0192] L is the last layer;

[0193] A (L) is the activation value of layer L;

[0194] g' (L)(Z (L) ) is the derivative of the activation function;

[0195] (W (l+1) ) T is the weight matrix;

[0196] δ (l+1) is the error term of the l+1th layer;

[0197] b. Weight gradient

[0198]

[0199] in:

[0200] is the weight gradient of the lth layer;

[0201] W (l) is the weight matrix;

[0202] A (l-1) is the activation value of the previous layer;

[0203] T is a mathematical symbol, meaning transpose;

[0204] c. Bias gradient

[0205]

[0206] in:

[0207] is the bias gradient of the lth layer;

[0208] b (l) is the bias vector;

[0209] is the error term of the lth layer;

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

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

[0212] The purpose of the previous steps is to identify cracks, but the number and shape of cracks in each image are different, so it is necessary to calculate indicators about cracks, namely the number, length, width, depth, maximum longitudinal extension length and maximum lateral extension length of cracks;

[0213] Step 11: Introduce industry experts to score fire risks based on the crack features of the image. The risk is divided into level 1 risk, level 2 risk, level 3 risk, and level 4 risk. The four risk levels correspond to extremely high risk, high risk, medium risk, and low risk, respectively. Combined with professional knowledge and experience, provide a high-level judgment on crack risks.

[0214] Step 12, using the number, length, width, depth, maximum longitudinal extension length and maximum transverse extension length of the cracks as inputs and expert scores as outputs, a decision tree model is established in the risk assessment module;

[0215] Step 13: For other underground coal-rock image data, the D-Net model network can be used to segment the cracks and calculate the crack index, and the crack index is input into the decision tree model of the risk assessment module to obtain the crack risk assessment result;

[0216] Through the above steps, the whole process from image input to fracture feature extraction, fracture location and risk assessment is realized, which improves the recognition accuracy of coal and rock fractures and the reliability of risk assessment;

[0217] Fire risk is scored based on the crack characteristics of the image. According to the characteristics of each stage of underground coal fire development under different conditions, the optimal combination and coordinated implementation plan of disaster prevention media suitable for it are studied, so as to study the precise optimization method of disaster prevention media based on the joint suppression of multi-physical field coupling process, build a theoretical and technical system for comprehensive prevention and control of underground coal fire, and realize precise targeted treatment of underground coal fire;

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

[0219] Level 1 risk uses water, inhibitors, slurry, liquid inert gas, gel, three-phase foam, and ultrafine water mist for comprehensive prevention and control of risk areas;

[0220] Level 2 risk uses water, slurry, inert gas, liquid inert gas, physical inhibitors, three-phase foam, and ultrafine water mist to prevent and control the risk area;

[0221] Level 3 risk uses three-phase foam, slurry, coal oxidation inhibitor, solidified foam, and colloid to prevent and control the risk area;

[0222] For level 4 risk, inert gas, liquid inert gas, three-phase foam and ultrafine water mist are used to prevent and control risk areas.

[0223] The above contents are only preferred embodiments of the present invention. For ordinary technicians in this field, many changes can be made in the specific implementation methods and application scopes based on the ideas of the present invention. As long as these changes do not deviate from the concept of the present invention, they all belong to the protection scope of the present invention.

Claims

1. The underground precise fire prevention and extinguishing coordinated system for mines based on rock mass fissure identification is characterized by: It includes data acquisition module, rock fracture identification module and risk assessment module; The data acquisition module is used to collect influencing factors related to coal-rock fracture development and underground coal-rock image data; The rock mass crack identification module is used to construct a D-Net model network to accurately locate the specific position of the crack in the image, and calculate the number, length, width, depth, maximum longitudinal extension length and maximum lateral extension length of the crack; The risk assessment module performs risk assessment on the cracks based on the results obtained by the rock crack identification module and the fire risk scores of industry experts based on the crack characteristics of the image.

2. According to the rock mass fissure identification-based precise underground mine fire prevention and extinguishing coordination system as described in claim 1, it is characterized by: The factors affecting the development of coal rock fractures 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.

3. According to the rock mass fissure identification-based precise underground mine fire prevention and extinguishing coordination system as described in claim 1, it is characterized by: The rock mass fracture identification module includes an Encoder module, a Decoder module, a DropBlock module, a DRB module and a feature fusion module; The Encoder module expresses scale features through pooling; The Decoder module expresses scale features through upsampling; The DropBlock module is used to process convolution features and suppress overfitting of convolutional neural networks; The DRB module selects the best path among multiple paths through learnable weights or routing functions according to the dynamic distribution of input features, while maintaining the global balance of the network; The feature fusion module is used to improve the expressiveness of the model.

4. A method for precise underground fire prevention and extinguishing in a mine based on rock mass fissure identification, using a precise underground fire prevention and extinguishing synergistic system for a mine based on rock mass fissure identification as claimed in any one of claims 1 to 3, specifically comprising the following steps: Step 1: collecting influencing factors related to coal-rock fracture development and underground coal-rock images through a data acquisition module; Step 2: Cut the underground coal and rock images to ensure the same size, and enhance the image binarization and adaptive histogram equalization to improve the clarity of the cracks, thereby improving the sensitivity and recognition accuracy of the rock crack recognition module to the crack characteristics; Step 3: Use image annotation software to annotate the cracks and divide the annotated data set into a training set and a test set; Step 4: Build a D-Net model network, which includes multiple levels of convolutional layers and pooling layers to extract features from images of different scales. Step 5: The D-Net model network includes an Encoder module and a Decoder module. Each module has a feature extraction structure of four scales. The Encoder module expresses scale features through pooling, and the Decoder module expresses scale features through upsampling. Step 6: Each feature extraction model contains a convolutional layer and a DropBlock module, and introduces a DRB module. This model makes full use of limited labeled data to train a deeper network and extract more complex features. Step 7: After extracting features at each level, the multi-scale features are integrated through the feature fusion module to form a more representative feature representation, combining low-level detail features with high-level semantic features to enhance the representation ability of features; Step 8: After feature fusion, use the convolutional layer to segment the image and calculate the loss for back propagation to accurately locate the specific location of the crack in the image. Step 9: After training the D-Net model network on the training set, the model weight parameters are obtained. The weights will be used to segment the cracks in the image during the model test. Step 10, calculating the number, length, width, depth, maximum longitudinal extension length and maximum transverse extension length of the cracks; Step 11: Introduce industry experts to score fire risks based on the crack features of the image. The risk is divided into level 1 risk, level 2 risk, level 3 risk, and level 4 risk. The four risk levels correspond to extremely high risk, high risk, medium risk, and low risk, respectively. Combined with professional knowledge and experience, provide a high-level judgment on crack risks. Step 12, using the number, length, width, depth, maximum longitudinal extension length and maximum transverse extension length of the cracks as inputs and expert scores as outputs, a decision tree model is established in the risk assessment module; Step 13. For other underground coal-rock image data, the D-Net model network can be used to segment the cracks and calculate the crack indicators. The crack indicators are input into the decision tree model of the risk assessment module to obtain the crack risk assessment results, and then precise targeted treatment of underground coal fires can be carried out.

5. The method for precise underground fire prevention and extinguishing in mines based on rock mass fissure identification according to claim 4 is characterized in that: In step six, the complete formula of the DropBlock module is expressed as: y n,c,h,w =x n,c,h,w ·M n,c,h,w , in: y n,c,h,w is the element of the output feature map; x n,c,h,w is the element of the input feature map; M n,c,h,w is the mask matrix after random generation and block operation; M is the mask matrix; p block Represents the total probability of the entire feature map being set to zero; The DropBlock module dynamically adjusts the drop probability p. During the training process, p gradually increases from a lower value to the target value p. max , the formula is: in: p max is the maximum drop probability; Current step is the current training step number; total steps is the total number of training steps; γ is a hyperparameter that controls the rate at which dropout increases.

6. The method for accurate fire prevention and extinguishing coordination in underground mines based on rock mass fissure identification according to claim 4 is characterized in that: In step six, the output of the DBR module is expressed as: At the same time, the optimization objective includes the loss function and the regularization term: in: Y is the output of the DBR module; z i is the activation value of path i; z j is the activation value of path j; P i (X) is the processing result of path i on input feature x; is the objective function term for optimization; Loss of the main task; To balance the hyperparameters, it is used to adjust the regularization strength.

7. The method for accurate fire prevention and extinguishing coordination in underground mines based on rock mass fissure identification according to claim 4 is characterized in that: In step seven, the general formula of the feature fusion module is expressed as: in: F out Fusion of multiple features into one output feature; F i is the i-th input feature map; n is the nth element; α i The weights are dynamically generated; Agg(F i ) represents the feature aggregation method.

8. The method for precise underground fire prevention and extinguishing coordination in mines based on rock mass fissure identification according to claim 4 is characterized in that: In step eight, the back propagation calculates the gradient of the loss function to the parameters layer by layer starting from the output layer through a recursive formula, and then updates the weights using an optimization algorithm. The core formula includes calculating the error term, weight gradient, and bias gradient using the chain rule. The recursive formula for back propagation is: a. Error term in: δ (l) is the error term of the lth layer; L is the last layer; A (L) is the activation value of layer L; g' (L) (Z (L) ) is the derivative of the activation function; (W (l+1) ) T is the weight matrix; δ (l+1) is the error term of the l+1th layer; b. Weight gradient in: is the weight gradient of the lth layer; W (l) is the weight matrix; A (l-1) is the activation value of the previous layer; c. Bias gradient in: is the bias gradient of the lth layer; b (l) is the bias vector; is the error term of the lth layer.

9. The method for precise underground fire prevention and extinguishing coordination in mines based on rock mass fissure identification according to claim 4 is characterized in that: In step thirteen, the underground coal fire is precisely targeted and treated as follows: Level 1 risk uses water, inhibitors, slurry, liquid inert gas, gel, three-phase foam, and ultrafine water mist for comprehensive prevention and control of risk areas; Level 2 risk uses water, slurry, inert gas, liquid inert gas, physical inhibitors, three-phase foam, and ultrafine water mist to prevent and control the risk area; Level 3 risk uses three-phase foam, slurry, coal oxidation inhibitor, solidified foam, and colloid to prevent and control the risk area; For level 4 risk, inert gas, liquid inert gas, three-phase foam and ultrafine water mist are used to prevent and control risk areas.

Images