Mine fire grading early warning method, system, equipment, medium and product

Abstract

The invention discloses a mine fire grading early warning method, system, device, medium and product, and the method comprises the steps: constructing an underground global map of a mine, and obtaining the multi-modal data of all local areas of the mine according to the underground global map; performing preprocessing and three-level data fusion on the multi-modal data, and determining a preliminary fire early warning result of the corresponding local area by adopting an edge hybrid neural network according to the fused multi-modal data; and according to the global multi-modal data of the mine, the initial fire early warning result and the real-time dynamic early warning threshold, performing fire risk prediction on the mine by adopting a central hybrid neural network to obtain a fire graded early warning result of the mine. The fire risk can be predicted more accurately based on multi-modal data fusion and a hybrid neural network; the early warning threshold value and strategy are automatically adjusted according to the real-time change of the mine environment by utilizing the technologies of online incremental learning, meta learning, reinforcement learning and the like, the parameters do not need to be manually adjusted, real-time calibration is realized, and the timeliness of early warning is improved.

Description

Technical Field

[0001] This invention relates to the field of mine fire technology, and in particular to a mine fire classification and early warning method, system, equipment, medium and product. Background Technology

[0002] Existing fire early warning methods mostly rely on fixed thresholds to trigger alarms (such as temperature and smoke concentration) and depend on data from a single sensor, failing to adequately adapt to the complex and dynamically changing environment of mines. Because they use static, fixed thresholds as alarm trigger conditions, they do not consider the dynamic differences in the mine environment (such as ventilation conditions and differences in work areas) to adjust the warning level, leading to high false alarm rates or missed alarm risks. For example, traditional systems do not distinguish between the warning differences in the initial smoldering and open flame stages, potentially delaying rescue efforts or causing over-response. Furthermore, because they rely heavily on single types of sensors (such as smoke and temperature), they lack collaborative analysis of multi-dimensional data such as gas concentration, gas drift, temperature and humidity interference, geological displacement, and personnel location. This results in a single warning dimension, failing to comprehensively reflect the fire risk status. Simultaneously, existing fire early warning methods mostly use static algorithm models based on fixed thresholds (such as CO concentration thresholds), lacking adaptability to dynamic changes in the mine environment. They cannot automatically optimize warning strategies based on changes in the mine environment (such as ventilation adjustments, impacts on mining progress, and equipment upgrades), requiring manual parameter adjustments, lacking real-time calibration, and exhibiting poor timeliness. Summary of the Invention

[0003] This invention provides a method, system, equipment, medium, and product for graded early warning of mine fires. Based on multimodal data fusion and hybrid neural networks, it can more accurately predict fire risks. By utilizing technologies such as online incremental learning, meta-learning, and reinforcement learning, it automatically adjusts the early warning threshold and strategy according to real-time changes in the mine environment, without relying on manual parameter adjustment, thereby achieving real-time calibration and improving the timeliness of early warnings.

[0004] To achieve the above objectives, embodiments of the present invention provide a method for graded early warning of mine fires, comprising: Construct a global underground map of the mine, and obtain multimodal data of all local areas of the mine based on the global underground map; Multimodal data is preprocessed and fused at three levels. Based on the fused multimodal data, an edge hybrid neural network is used to determine the preliminary fire warning results for the corresponding local area. Based on the mine's global multimodal data, preliminary fire early warning results, and real-time dynamic early warning thresholds, a central hybrid neural network is used to predict the mine's fire risk and obtain the mine's fire classification early warning results.

[0005] As an improvement to the above scheme, the construction of an underground global map of the mine, and the acquisition of multimodal data of all local areas of the mine based on the underground global map, includes: A mobile inspection robot was used to collect laser point cloud data and IMU attitude data for all local areas of the mine. Based on laser point cloud data and IMU attitude data, an improved Cartographer algorithm is used for simultaneous localization and map building to generate local maps of corresponding local areas. By merging and stitching together all the local maps, a global underground map of the mine can be obtained; Mobile inspection robots and fixed sensor networks are used to acquire multimodal data of all local areas in the mine.

[0006] As an improvement to the above scheme, the preprocessing and three-level data fusion of multimodal data, and the determination of preliminary fire early warning results for corresponding local areas using an edge hybrid neural network based on the fused multimodal data, include: Edge servers are used to perform data cleaning, standardization, and time-series alignment on multimodal data; A three-level fusion architecture is adopted to perform sensor-level fusion, feature-level fusion, and decision-level fusion on the processed multimodal data; Based on the fused multimodal data, an edge hybrid neural network is used to determine the preliminary fire warning results for the corresponding local area.

[0007] As an improvement to the above scheme, the three-level fusion architecture for sensor-level fusion, feature-level fusion, and decision-level fusion of the processed multimodal data includes: An improved Kalman filtering algorithm is used to calibrate and synchronize sensor data of the same type in the processed multimodal data to obtain single-type accurate data; A deep cross-network is used to model the interaction relationships between different features in single-type precise data to generate feature vectors; Based on the feature vectors, the prediction tendency of the corresponding multi-model in the edge hybrid neural network is synthesized using DS evidence theory. A conflict factor is introduced to handle the conflict of prediction tendency, and the target feature vector corresponding to the low conflict prediction tendency is obtained.

[0008] As an improvement to the above scheme, the edge hybrid neural network and the center hybrid neural network are continuously learned and adaptively optimized through an adaptive learning framework; The edge hybrid neural network includes a Transformer module and a bidirectional gated recurrent unit; The Transformer module includes a multi-head self-attention layer, positional encoding, and a forward propagation network.

[0009] As an improvement to the above scheme, the real-time dynamic early warning threshold is adjusted using an adaptive threshold and a dynamic modeling module.

[0010] To achieve the above objectives, embodiments of the present invention provide a mine fire classification and early warning system, comprising: The data acquisition layer, consisting of a fixed sensor network and multiple mobile inspection robots, is used to construct a global underground map of the mine and acquire multimodal data of all local areas of the mine based on the global underground map. The edge processing layer includes edge servers deployed at various key nodes underground, used for preprocessing multimodal data and three-level data fusion. Based on the fused multimodal data, an edge hybrid neural network is used to determine the preliminary fire warning results for the corresponding local area. The cloud-based decision-making layer, including cloud servers, communicates with various edge servers to predict the fire risk of the mine using a central hybrid neural network based on the mine's global multimodal data, preliminary fire warning results, and real-time dynamic warning thresholds, thereby obtaining the mine's fire classification warning results.

[0011] To achieve the above objectives, this invention provides a mine fire classification and early warning device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the above-mentioned mine fire classification and early warning method.

[0012] To achieve the above objectives, embodiments of the present invention also provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the above-described mine fire classification and early warning method.

[0013] To achieve the above objectives, embodiments of the present invention also provide a computer program product, which is stored in a storage medium and executed by at least one processor to implement the steps of the above-described mine fire classification and early warning method.

[0014] Compared with existing technologies, the present invention discloses a mine fire classification and early warning method, system, equipment, medium, and product. This method constructs a global underground map of the mine and obtains multimodal data of all local areas based on the map. The multimodal data undergoes preprocessing and three-level data fusion. Based on the fused multimodal data, an edge hybrid neural network is used to determine the preliminary fire warning result for the corresponding local area. Based on the global multimodal data of the mine, the preliminary fire warning result, and the real-time dynamic warning threshold, a central hybrid neural network is used to predict the fire risk of the mine, obtaining the mine's fire classification and early warning result. The multimodal data fusion and hybrid neural network enable more accurate fire risk prediction. Utilizing online incremental learning, meta-learning, and reinforcement learning technologies, the warning threshold and strategy are automatically adjusted according to real-time changes in the mine environment, eliminating the need for manual parameter adjustments and achieving real-time calibration and improved warning timeliness. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a mine fire classification and early warning method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a mine fire classification and early warning system provided in an embodiment of the present invention; Figure 3 This is a structural block diagram of a mine fire classification and early warning device provided in an embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] It should be noted that the terms "comprising" and "specific" in this invention, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0018] Please see Figure 1 , Figure 1 This is a flowchart illustrating a mine fire classification and early warning method provided in an embodiment of the present invention. The mine fire classification and early warning method includes: S1, Construct a global underground map of the mine, and obtain multimodal data of all local areas of the mine based on the global underground map; S2, preprocesses and fuses multimodal data at three levels, and uses an edge hybrid neural network to determine the preliminary fire warning result for the corresponding local area based on the fused multimodal data; S3. Based on the mine's global multimodal data, preliminary fire warning results, and real-time dynamic warning thresholds, a central hybrid neural network is used to predict the mine's fire risk and obtain the mine's fire classification warning results.

[0019] For example, the mobile inspection robot initializes SLAM, loads a pre-stored tunnel topology map, activates tight-coupled positioning of LiDAR and IMU, performs autonomous inspection, and constructs a local map of its own area. The mobile inspection robot uploads its local map to the cloud server via an edge server, and constructs a global underground map through point cloud stitching. Multimodal data is collected using a fixed sensor network and the mobile inspection robot, and uploaded to the edge server. This multimodal data includes environmental parameters, gas monitoring, visual perception, and physical feature data. The edge server preprocesses the collected data and performs three-level fusion, then performs real-time early warning judgment on the local data, initially assesses fire risk, and uploads the warning results to the cloud server. The cloud server utilizes a T-based... The ransformer module, combined with a bidirectional GRU hybrid neural network module, performs comprehensive analysis of global multimodal data, extracting long-term dependent features and temporal variation information to predict fire risk levels and trends. The cloud server continuously learns and adaptively optimizes the hybrid neural network mounted on the cloud server and edge server through a continuous adaptive learning framework. The cloud server dynamically adjusts the threshold of the hybrid neural network through adaptive thresholding and dynamic modeling modules, ensuring the system's adaptability to different scenarios. The cloud server, through its spatial positioning and fire tracking module, integrates early warning signals from various regions and global map information, using graph neural networks to analyze the mine's spatial structure, predict fire spread paths and ignition point locations, and generate a global fire risk classification early warning result. Fixed sensor networks consist of the following three types of sensors: Environmental parameter sensors include: Temperature sensor: Uses thermocouple or infrared temperature measurement module, sampling frequency 1Hz, measurement range -20℃~1000℃, accuracy ±0.5℃.

[0020] Humidity sensor: Capacitive sensor, range 0-100%RH, accuracy ±2%RH.

[0021] Wind speed sensor: Ultrasonic anemometer, based on time-of-flight method to measure wind speed, range 0-30m / s, resolution 0.1m / s.

[0022] Gas monitoring sensors include: CO sensor: electrochemical principle, range 0-1000ppm, response time <30s.

[0023] CH4 sensor: infrared absorption principle, measurement range 0-100%LEL, resistant to H2O and CO2 interference.

[0024] Physical feature sensors: Vibration sensor: piezoelectric accelerometer, range ±50g, detects abnormal vibration signals of the equipment (frequency range 0.5-5kHz).

[0025] Acoustic sensor: MEMS microphone array, signal-to-noise ratio >64dB.

[0026] The environmental parameter data collected by the above sensors includes temperature, humidity, wind speed, air pressure, and dust concentration; gas monitoring data includes CO, CO2, O2, and methane concentrations and their rate of change; visual perception data includes thermal imaging data and high-definition video streams; and physical characteristic data includes vibration and sound.

[0027] The visual sensors mounted on the mobile inspection robot include LiDAR, inertial measurement unit, binocular camera, and thermal imager; Environmental and gas detection sensors include sensors for temperature and humidity, air pressure, wind speed, dust concentration, smoke concentration, methane concentration, CO concentration, CO2 concentration, O2 concentration, and acoustic vibration signals.

[0028] Specifically, the construction of the underground global map of the mine, and the acquisition of multimodal data of all local areas of the mine based on the underground global map, include: A mobile inspection robot was used to collect laser point cloud data and IMU attitude data for all local areas of the mine. Based on laser point cloud data and IMU attitude data, an improved Cartographer algorithm is used for simultaneous localization and map building to generate local maps of corresponding local areas. By merging and stitching together all the local maps, a global underground map of the mine can be obtained; Mobile inspection robots and fixed sensor networks are used to acquire multimodal data of all local areas in the mine.

[0029] For example, the mobile inspection robot's LiDAR uses a 16-line radar with a scanning frequency of 10Hz and a detection range of 50m. Combined with an improved Cartographer algorithm, it achieves SLAM mapping with a positioning error of <5cm. Its inertial measurement unit integrates a gyroscope (range ±2000° / s) and an accelerometer (range ±16g), outputting 6-DOF attitude data. Using LiDAR and visual information, a local map is constructed using a graph optimization-based SLAM algorithm (such as Graph-SLAM or Cartographer). The graph optimization objective function is: , in, For the observations between node i and node j, For the prediction function, To observe the noise covariance matrix.

[0030] Specifically, the preprocessing and three-level data fusion of multimodal data, and the determination of preliminary fire early warning results for corresponding local areas using an edge hybrid neural network based on the fused multimodal data, include: Edge servers are used to perform data cleaning, standardization, and time-series alignment on multimodal data; A three-level fusion architecture is adopted to perform sensor-level fusion, feature-level fusion, and decision-level fusion on the processed multimodal data; Based on the fused multimodal data, an edge hybrid neural network is used to determine the preliminary fire warning results for the corresponding local area.

[0031] Specifically, the three-level fusion architecture for performing sensor-level fusion, feature-level fusion, and decision-level fusion on the processed multimodal data includes: An improved Kalman filtering algorithm is used to calibrate and synchronize sensor data of the same type in the processed multimodal data to obtain single-type accurate data; A deep cross-network is used to model the interaction relationships between different features in single-type precise data to generate feature vectors; Based on the feature vectors, the prediction tendency of the corresponding multi-model in the edge hybrid neural network is synthesized using DS evidence theory. A conflict factor is introduced to handle the conflict of prediction tendency, and the target feature vector corresponding to the low conflict prediction tendency is obtained.

[0032] For example, the edge server processes data collected by fixed sensor networks and mobile inspection robots as follows: It constructs a three-tier fusion architecture to achieve deep data integration, including: Sensor-level fusion: At the sensor level, an improved Kalman filtering algorithm is used to calibrate and synchronize multi-sensor data; the state equation of the improved adaptive Kalman filter is as follows: , The measurement equation is: , Where A is the state transition matrix, B is the control matrix, H is the observation matrix, and w and v are Gaussian noise.

[0033] Feature-level fusion: At the feature level, a deep cross-network (DCN) is used to model the interaction relationships between different features. , in, For the original input, These are learnable weights. The number of cross layers is L=6, and the parameters are initialized using the Xavier method.

[0034] Decision-level fusion: At the decision level, multi-model outputs are synthesized using DS evidence theory, and a conflict factor is introduced to handle prediction conflicts between different models, thereby improving fusion reliability. , , in, and These are the evidence distributions for the Transformer and BiGRU models, respectively, with K being the conflict factor.

[0035] Principal component analysis is used for feature dimensionality reduction.

[0036] Specifically, the edge hybrid neural network and the center hybrid neural network are continuously learned and adaptively optimized through an adaptive learning framework; The edge hybrid neural network includes a Transformer module and a bidirectional gated recurrent unit; The Transformer module includes a multi-head self-attention layer, positional encoding, and a forward propagation network.

[0037] For example, a hybrid neural network includes a Transformer module and a bidirectional gated recurrent unit (BiGRU). The Transformer is responsible for capturing long-term dependencies between multiple parameters, including: Multi-head self-attention layer: used to automatically learn the correlation between data from different modalities, represented as: , Where Q, K, and V are query, key, and value matrices, with 8 heads.

[0038] Location coding: used to preserve the spatial distribution information of the sensor; to preserve the location information of each data point in the sequence, location codes are generated using sine and cosine functions. , , Forward propagation network: used to generate high-dimensional feature vectors, representing intermediate representations of the fire state; includes two linear transformations and one non-linear activation function (e.g., ReLU). , The Bi-Gated Cyclic Unit (BiGRU) is used to process time-series data and predict the development trend of a fire. For each time t, the forward GRU update formula is: , The hidden layer has a dimension d=256 and a time step T=60. The bidirectional structure considers both forward and backward information, enhancing the representation of temporal features.

[0039] Specifically, the real-time dynamic early warning threshold is adjusted using an adaptive threshold and a dynamic modeling module.

[0040] The cloud server includes a continuous adaptive learning framework, an adaptive threshold and dynamic modeling module, and a spatial positioning and fire tracking module. The continuous adaptive learning framework is used to continuously learn and adaptively optimize hybrid neural networks mounted on cloud servers and edge servers. The adaptive threshold and dynamic modeling module is used to realize dynamic threshold adjustment of hybrid neural networks, ensuring the system's adaptability to different scenarios; The spatial positioning and fire tracking module stitches and deduplicates the SLAM local map constructed by the mobile inspection robot to obtain the underground global map; it uses graph neural network (GNN) to analyze spatial data and combines it with temporal data to predict the fire spread path.

[0041] Optional, continuous adaptive learning frameworks include: Incremental learning and online learning modules: These enable the system to continuously update model parameters based on real-time data, thereby adapting to dynamic changes in the mine environment and reducing the risk of false alarms and missed alarms. The incremental learning strategy is as follows: design an initial pre-trained model as a foundation; when the system collects new data during actual operation, it performs small-batch online fine-tuning, updating only the last few layers of the model to reduce the destruction of existing knowledge. The online learning strategy is as follows: adopt algorithms based on online gradient descent or adaptive learning rate to update model parameters in real time on edge computing devices; periodically use newly collected data for short-cycle training, and save the incremental update log of the model for traceability; Model Adaptation Mechanism Module: This module designs a fast adapter based on Model-Agnostic Meta-Learning (MAML) to enable the model to quickly adapt to new environments, allowing the system to converge and adapt rapidly with only a small amount of data in new environments. The outer layer update formula in MAML is: , The inner layer is updated as follows: , Where α and β are the learning rates of the inner and outer layers, respectively. Contrastive learning module: Constructs a triplet loss function so that the distance between the anchor point and the positive sample is smaller than the distance between the anchor point and the negative sample, thereby optimizing the model's metrics.

[0042] Positive samples p come from historical data under the same working conditions, and negative samples n come from different working conditions. Margin=0.5. This method enhances the discriminative power of the features. The reinforcement learning optimization module employs a deep Q-network (DQN) to dynamically optimize the early warning strategy, which includes: State space: contains sensor data, environmental parameters, and system state; Action space: including adjusting early warning thresholds, allocating inspection resources, and adjusting monitoring frequency. Reward function: Combining early warning accuracy, timeliness, and resource consumption; By interacting and learning with the environment, the system can autonomously discover the optimal early warning strategy, achieving the goal of "fewer false alarms and fewer missed alarms".

[0043] Optional, the adaptive thresholding and dynamic modeling module includes: An online learning algorithm is used to adjust fire warning thresholds based on real-time data. First, an initial threshold is set based on historical data. Then, by inputting real-time sensor data, a dynamic threshold model is constructed using Gaussian process regression to automatically adjust the threshold according to the coal quality and geological conditions of different mines. The warning sensitivity is dynamically adjusted considering historical data trends. A sliding window is introduced to avoid drastic threshold fluctuations. Assuming the observed data is... Where yi represents the early warning trigger index, the mean value predicted by the Gaussian process regression is: , in, For kernel functions, the RBF kernel is often chosen: , The threshold is smoothed by combining a sliding window.

[0044] Based on digital twin technology, a digital twin of the mine is constructed, and the mine fire model is updated in real time. In the digital twin of the mine, geological data, ventilation network and equipment layout are integrated to build a physical model. The model parameters are continuously corrected using real-time data. The fire development trend is simulated through computational fluid dynamics (CFD). The optimal evacuation route is dynamically calculated. The warning level is dynamically adjusted according to the simulation results. In this embodiment, a simplified CFD model is established to solve the Navier-Stokes equations and the energy transfer equations: , Where u is the velocity vector, p is the pressure, ρ is the density, ν is the kinematic viscosity, T is the temperature, α is the thermal diffusivity, and Q is the heat source term.

[0045] Combining domain expert knowledge with a data-driven model, a knowledge graph for the mine fire domain is constructed, encoding expert experience. A rule-based inference engine is designed, and a rule-constrained loss function is implemented. Expert knowledge is used to constrain the deep learning model, improving interpretability. Data-driven and knowledge-driven results are mutually validated. The knowledge graph includes nodes (sensors, equipment, regions) and edges (physical connections, environmental relationships). Graph reasoning is implemented using graph databases such as Neo4j, and the rule-constrained loss function is designed as follows: , in, Data-driven loss (such as cross-entropy, mean squared error). Here, λ represents the constraint term based on expert rules, and λ is the balance coefficient.

[0046] Optional, the spatial positioning and fire tracking module includes: By using multi-point gas sensor data, the concentration gradient is calculated to infer the ignition point direction, outputting the gas diffusion path and possible ignition area; acoustic vibration signal sensors are used to capture fire explosion or combustion sounds, and the ignition point is located by combining the sound wave propagation time difference, outputting the three-dimensional coordinates of the ignition point; a weighted fusion algorithm is used to fuse the above results; and the gas concentration gradient is used... C analyzes the direction of the fire source; the basic formula is: , Using graph neural networks (GNNs) to analyze spatial data and predict fire spread paths, specifically including: The graph data structure and multi-source localization results are used as input. The graph data structure includes: mine roadways as nodes, connections as edges, and node attributes including multi-dimensional data collected by fixed sensor networks and mobile inspection robots. A graph G=(V,E) is generated based on the mine topology. Node features include multi-dimensional data such as temperature, gas concentration, and humidity. The GNN analyzes the spatial relationships between adjacent nodes through a message passing mechanism, described as follows: , Where σ is the ReLU activation function, This represents the feature vector of node v in the 111 layer. Let v be the set of its neighboring nodes. and For learnable weights and biases.

[0047] By combining wind speed and the tunnel structure displayed on the global map, the direction of fire spread is predicted, and the coordinates of the ignition point and the path of fire spread are finally output.

[0048] Optionally, the fire classification early warning results include classification early warnings in both spatial and temporal dimensions; Spatial hierarchical early warning is a hierarchical early warning system that extends from the entire mine to the working face, and then to specific locations, including: Mine level: Full mine risk assessment; Full mine risk index > 0.7, triggering full mine evacuation.

[0049] Roadway level: Risk at the working face or mining area; local CO concentration >100ppm and temperature >60℃, the roadway shall be sealed off.

[0050] Equipment level: Risks at specific equipment or locations; if the equipment vibration frequency is >3kHz, shut off the power and dispatch a robot for verification.

[0051] Time-based early warning refers to early warning on a time scale that ranges from long-term trends to medium-term changes and then to short-term rapid changes. It includes long-term trend prediction, medium-term early warning, and immediate early warning achieved by hybrid neural networks.

[0052] This invention discloses a method for graded early warning of mine fires. It constructs a global underground map of the mine and obtains multimodal data of all local areas based on this map. The multimodal data undergoes preprocessing and three-level data fusion. Based on the fused multimodal data, an edge hybrid neural network is used to determine the preliminary fire warning result for the corresponding local area. Based on the global multimodal data of the mine, the preliminary fire warning result, and the real-time dynamic warning threshold, a central hybrid neural network is used to predict the fire risk of the mine, resulting in a graded early warning result. The method, based on multimodal data fusion and hybrid neural networks, can more accurately predict fire risk. Utilizing online incremental learning, meta-learning, and reinforcement learning techniques, it automatically adjusts the warning threshold and strategy according to real-time changes in the mine environment, eliminating the need for manual parameter adjustments and achieving real-time calibration and improved warning timeliness.

[0053] See Figure 2 , Figure 2 This is a schematic diagram of the structure of a mine fire classification and early warning system 10 provided in an embodiment of the present invention. The mine fire classification and early warning system 10 includes: The data acquisition layer 11 consists of a fixed sensor network and multiple mobile inspection robots, which are used to construct a global underground map of the mine and acquire multimodal data of all local areas of the mine based on the global underground map. The edge processing layer 12 includes edge servers deployed at various key nodes underground, used for preprocessing multimodal data and three-level data fusion, and using edge hybrid neural networks to determine the preliminary fire warning results for the corresponding local area based on the fused multimodal data; The cloud-based decision layer 13, including cloud servers, communicates with various edge servers to predict the fire risk of the mine based on the mine's global multimodal data, preliminary fire warning results, and real-time dynamic warning thresholds, using a central hybrid neural network to obtain the mine's fire classification warning results.

[0054] The mine fire classification and early warning system 10 provided in this embodiment of the invention can realize all the processes of the mine fire classification and early warning method of the above embodiment. The functions and technical effects of each module in the system are the same as those of the mine fire classification and early warning method of the above embodiment, and will not be repeated here.

[0055] See Figure 3 , Figure 3 This is a schematic diagram of the structure of a mine fire classification and early warning device 20 provided in an embodiment of the present invention. The mine fire classification and early warning device 20 of this embodiment includes: a processor 21, a memory 22, and a computer program stored in the memory 22 and executable on the processor 21. When the processor 21 executes the computer program, it implements the steps in the above-described mine fire classification and early warning method embodiment. Alternatively, when the processor 21 executes the computer program, it implements the functions of each module in the above-described mine fire classification and early warning system embodiment.

[0056] For example, the computer program may be divided into one or more modules, which are stored in the memory 22 and executed by the processor 21 to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the mine fire classification and early warning device 20.

[0057] The mine fire classification and early warning device 20 can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. The mine fire classification and early warning device 20 may include, but is not limited to, a processor 21 and a memory 22. Those skilled in the art will understand that the schematic diagram is merely an example of the mine fire classification and early warning device 20 and does not constitute a limitation on the device. It may include more or fewer components than shown, or combine certain components, or use different components. For example, the mine fire classification and early warning device 20 may also include input / output devices, network access devices, buses, etc.

[0058] The processor 21 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor 21 is the control center of the mine fire classification and early warning device 20, connecting all parts of the device through various interfaces and lines.

[0059] The memory 22 can be used to store the computer programs and / or modules. The processor 21 implements various functions of the mine fire classification and early warning device 20 by running or executing the computer programs and / or modules stored in the memory 22 and calling the data stored in the memory 22. The memory 22 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 22 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0060] The integrated modules of the mine fire classification and early warning equipment 20, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by the processor 21, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or system capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content contained in the computer-readable medium may be appropriately added to or subtracted from the content as required by the legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium may not include electrical carrier signals and telecommunication signals.

[0061] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0062] This invention also provides a computer-readable storage medium, which includes a stored computer program, wherein the computer program, when running, controls the device where the computer-readable storage medium is located to execute the mine fire classification and early warning method as described above.

[0063] Furthermore, embodiments of the present invention also provide a computer program product, which is stored in a storage medium and executed by at least one processor to implement the steps of the mine fire classification and early warning method described above.

[0064] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for graded early warning of mine fires, characterized in that, include: Construct a global underground map of the mine, and obtain multimodal data of all local areas of the mine based on the global underground map; Multimodal data is preprocessed and fused at three levels. Based on the fused multimodal data, an edge hybrid neural network is used to determine the preliminary fire warning results for the corresponding local area. Based on the mine's global multimodal data, preliminary fire early warning results, and real-time dynamic early warning thresholds, a central hybrid neural network is used to predict the mine's fire risk and obtain the mine's fire classification early warning results.

2. The mine fire classification and early warning method as described in claim 1, characterized in that, The construction of the underground global map of the mine involves obtaining multimodal data of all local areas of the mine based on the underground global map, including: A mobile inspection robot was used to collect laser point cloud data and IMU attitude data for all local areas of the mine. Based on laser point cloud data and IMU attitude data, an improved Cartographer algorithm is used for simultaneous localization and map building to generate local maps of corresponding local areas. By merging and stitching together all the local maps, a global underground map of the mine can be obtained; Mobile inspection robots and fixed sensor networks are used to acquire multimodal data of all local areas in the mine.

3. The mine fire classification and early warning method as described in claim 1, characterized in that, The process of preprocessing and three-level data fusion of multimodal data, and determining the preliminary fire early warning result for the corresponding local area using an edge hybrid neural network based on the fused multimodal data, includes: Edge servers are used to perform data cleaning, standardization, and time-series alignment on multimodal data; A three-level fusion architecture is adopted to perform sensor-level fusion, feature-level fusion, and decision-level fusion on the processed multimodal data; Based on the fused multimodal data, an edge hybrid neural network is used to determine the preliminary fire warning results for the corresponding local area.

4. The mine fire classification and early warning method as described in claim 3, characterized in that, The three-level fusion architecture is used to perform sensor-level fusion, feature-level fusion, and decision-level fusion on the processed multimodal data, including: An improved Kalman filtering algorithm is used to calibrate and synchronize sensor data of the same type in the processed multimodal data to obtain single-type accurate data; A deep cross-network is used to model the interaction relationships between different features in single-type precise data to generate feature vectors; Based on the feature vectors, the prediction tendency of the corresponding multi-model in the edge hybrid neural network is synthesized using DS evidence theory. A conflict factor is introduced to handle the conflict of prediction tendency, and the target feature vector corresponding to the low conflict prediction tendency is obtained.

5. The mine fire classification and early warning method as described in claim 1, characterized in that, The edge hybrid neural network and the center hybrid neural network are continuously learned and adaptively optimized through an adaptive learning framework; The edge hybrid neural network includes a Transformer module and a bidirectional gated recurrent unit; The Transformer module includes a multi-head self-attention layer, positional encoding, and a forward propagation network.

6. The mine fire classification and early warning method as described in claim 1, characterized in that, The real-time dynamic early warning threshold is adjusted using an adaptive threshold and a dynamic modeling module.

7. A mine fire classification and early warning system, characterized in that, include: The data acquisition layer, consisting of a fixed sensor network and multiple mobile inspection robots, is used to construct a global underground map of the mine and acquire multimodal data of all local areas of the mine based on the global underground map. The edge processing layer includes edge servers deployed at various key nodes underground, used for preprocessing multimodal data and three-level data fusion. Based on the fused multimodal data, an edge hybrid neural network is used to determine the preliminary fire warning results for the corresponding local area. The cloud-based decision-making layer, including cloud servers, communicates with various edge servers to predict the fire risk of the mine using a central hybrid neural network based on the mine's global multimodal data, preliminary fire warning results, and real-time dynamic warning thresholds, thereby obtaining the mine's fire classification warning results.

8. A mine fire classification and early warning device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the mine fire classification and early warning method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the mine fire classification and early warning method as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product is stored in a storage medium, and the program product is executed by at least one processor to implement the steps of the mine fire classification and early warning method as described in any one of claims 1-6.

Images