A method for real-time monitoring and intelligent early warning of mine disasters in well mines based on multi-sensor fusion

Abstract

The application discloses a kind of well mine disaster real-time monitoring and intelligent early warning method based on multi-sensor fusion, it is related to mine safety monitoring technical field, comprising the following steps: multi-source heterogeneous data acquisition and fusion pretreatment, double-layer safety state evaluation model construction and weight dynamic assignment, disaster risk real-time comprehensive evaluation and early warning classification, intelligent early warning measure matching and push, early warning effect feedback and model self-adaptive optimization, three-dimensional visual comprehensive display and situation deduction;The application integrates underground sensor and ground remote sensing system, collects multi-disaster data and time-space registration pretreatment into unified data stream, constructs double-layer evaluation model, dynamically assigns weight and introduces disaster coupling amplification coefficient quantitative correlation risk, generates structured early warning push after real-time evaluation intelligent matching rule base, optimizes model through early warning feedback reinforcement learning, three-dimensional visual display situation, significantly improve monitoring comprehensiveness, evaluation accuracy, early warning timeliness and disposal pertinence.

Description

Technical Field

[0001] This invention relates to the field of mine safety monitoring technology, and in particular to a method for real-time monitoring and intelligent early warning of underground mine disasters based on multi-sensor fusion. Background Technology

[0002] Mining is an industrial site for extracting useful minerals from the Earth's surface or crust. It is mainly divided into two categories: open-pit mining and underground mining (i.e., shaft mining). Underground mining refers to the operation of mining deep ore layers by excavating shafts and underground tunnel systems. Underground mining is typically used when the ore body is deeply buried or when open-pit mining is uneconomical. Its core facilities include hoisting shafts, ventilation shafts, haulage tunnels, working faces, and safety systems. The mining process requires the comprehensive application of technologies such as tunneling, support, ventilation, drainage, and gas control to ensure safety and efficiency. Underground mining operates in a complex environment, facing risks such as ground pressure, gas, and water hazards. It places extremely high demands on technical equipment and safety management, and is the main way to obtain deep mineral resources such as coal and metals.

[0003] Underground mines, due to their deep underground location and complex geological environment, face a variety of disaster threats. Major hazards include gas (methane, etc.) explosions and outbursts, mine water inrush and sudden water surges, rock bursts, roof collapses, coal dust explosions, and underground fires. These hazards are often sudden, cascading, and extremely destructive, seriously threatening the lives of miners and the safety of mine facilities. To ensure the safety of miners and mine facilities, real-time monitoring and early warning of underground mine hazards are necessary to achieve safe production in underground mines.

[0004] Based on existing technologies, current underground mine disaster monitoring and early warning technologies largely rely on single-sensor monitoring, resulting in problems such as heterogeneous data that is difficult to integrate, inconsistent spatiotemporal benchmarks, and the use of static weights in disaster assessment models, which cannot adapt to changes in mining stages and geological conditions, easily leading to inaccurate early warning results. Furthermore, traditional early warning technologies rely heavily on human experience and lack multi-hazard coupling analysis and intelligent matching mechanisms, as well as a feedback loop, making it difficult to iteratively optimize the model. This results in low monitoring accuracy, delayed early warning, and insufficient targeted response, making it difficult to cope with complex disaster scenarios. Therefore, this invention proposes a real-time monitoring and intelligent early warning method for underground mine disasters based on multi-sensor fusion to address the problems existing in the current technologies. Summary of the Invention

[0005] To address the aforementioned problems, the present invention aims to propose a real-time monitoring and intelligent early warning method for underground mine disasters based on multi-sensor fusion. This method solves the problems that existing underground mine disaster monitoring and early warning technologies rely heavily on single-sensor monitoring, disaster assessment models often use static weights, lack multi-hazard coupling analysis and intelligent matching mechanisms, and lack effective feedback loops, making it difficult to iteratively optimize the models.

[0006] To achieve the objectives of this invention, the invention is implemented through the following technical solution: a method for real-time monitoring and intelligent early warning of underground mine disasters based on multi-sensor fusion, comprising the following steps: Step 1: Collect multi-source heterogeneous real-time monitoring data related to underground mine disasters through underground sensor networks and surface remote sensing monitoring systems, perform data fusion preprocessing, generate a standardized data stream with unified spatiotemporal characteristics, and upload it to the integrated management and control platform; Step 2: Combining the Analytic Hierarchy Process (AHP) with expert experience, a two-layer safety status evaluation model is constructed within the integrated management and control platform, consisting of a safety status evaluation model for several disaster subsystems at the lower level and a safety status evaluation model for the overall mine at the upper level. Historical data and expert questionnaires are used to dynamically assign weights. Step 3: Input the standardized data stream into the two-layer safety status evaluation model in real time. The safety status evaluation models of each disaster subsystem calculate and output scores and risk levels in parallel, and input them into the overall mine safety status evaluation model to obtain the overall real-time comprehensive safety status score and risk level of the mine. Step 4: When the risk level reaches the threshold, the integrated management and control platform triggers an early warning, intelligently matching and generating structured information containing disposal suggestions, responsible personnel, and execution procedures from the expert rule base according to the disaster type and level, and pushing it through multiple channels; Step 5: The integrated management and control platform records the data of early warning events, the implementation of measures and the effects of handling them, and regularly uses reinforcement learning to iteratively optimize the dynamic weighting parameters in Step 2, so that the overall safety status evaluation model of the mine can adapt to changes in mining stages and geological conditions. Step Six: In the integrated 3D geological and roadway model of the mine integrated management and control platform, dynamically display the multi-source heterogeneous real-time monitoring data collected in Step One, the heat map of risk level output in Step Three, the early warning information triggered in Step Four, and short-term simulation of disaster trends.

[0007] Further improvements are made in the following aspects: In step one, the downhole sensor network includes hydrological monitoring sensors, temperature sensors, gas monitoring sensors, dust concentration sensors, roof delamination meters, and anchor bolt and cable stress sensors; the surface remote sensing monitoring system includes UAV remote sensing equipment and satellite remote sensing equipment; the multi-source heterogeneous real-time monitoring data includes data related to water hazards, fires, gas disasters, dust disasters, roof disasters, mine pressure disasters, and geological disasters; and the data fusion preprocessing includes timestamp alignment, spatial location registration, format standardization, and outlier cleaning.

[0008] Further improvements are made in the following ways: the spatial registration establishes a three-dimensional mapping relationship between all sensor data and the unified geographic coordinate system of the mine, so that each data point is associated with the location of the underground roadway or the surface coordinates. The geological disaster-related data monitoring integrates UAV visible light and infrared image data and satellite synthetic aperture radar interferometry data, and quantitatively obtains surface subsidence and ground fissure displacement field information through image processing and deformation inversion, as part of the standardized data stream.

[0009] A further improvement lies in the following: In step two, the specific steps for dynamically assigning weights are as follows: a) A group of experts constructs a judgment matrix by comparing each pair of evaluation indicators at the same level; b. Examine and correct the judgment matrix; c. Integrate the judgment matrix that has passed the test with the objective weighting factors of historical data statistics to form a comprehensive weight.

[0010] A further improvement is made in step two, where a disaster coupling amplification coefficient is introduced to weight the overall mine evaluation model. This disaster coupling amplification coefficient is derived from the analysis of the chain-like symbiosis and aggravation relationship of multiple disasters in historical accidents and is used to correct the weights of subsystems to quantify the associated risks of disasters.

[0011] A further improvement is that, in step three, the risk level is divided into four levels: safe, caution, warning, and danger, with the threshold determined comprehensively based on industry standards, mine geological conditions, and historical data distribution.

[0012] A further improvement is made in step four, where the expert rule base uses ontology knowledge representation to formally define and store disaster entities, monitoring indicators, risk status, response actions, resource objects and relationships, forming a reasonable knowledge network.

[0013] Further improvements are made in the following steps: In step four, intelligent matching generates structured information by performing graph traversal and rule reasoning based on a reasonable knowledge network, and combining basic handling rules to form an operable comprehensive plan that adapts to complex scenarios.

[0014] The beneficial effects of this invention are as follows: By integrating underground sensor networks and surface remote sensing systems, this invention synchronously collects multi-hazard data and generates a unified data stream through data fusion preprocessing, thus solving the fragmentation problem of traditional single-sensor monitoring. On this basis, a two-layer evaluation model is constructed, combining the analytic hierarchy process, expert experience, and dynamic weighting of historical data. A disaster coupling amplification coefficient is introduced to quantify the associated risks of multiple hazards, improving the accuracy of the evaluation. Furthermore, after real-time evaluation, the system intelligently matches the expert rule base to generate structured early warnings containing disposal suggestions and pushes them through multiple channels, enhancing the targeted nature of the response. At the same time, the model is iteratively optimized using reinforcement learning based on early warning feedback, combined with three-dimensional visualization to dynamically display the risk situation. Ultimately, this invention achieves dynamic perception of disaster risks, adaptive prevention and control, and intelligent protection across the entire chain, significantly improving the comprehensiveness and reliability of underground mine safety monitoring. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the real-time monitoring and intelligent early warning method for underground mine disasters based on multi-sensor fusion, as described in this 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] The integrated management and control platform in this embodiment serves as the core carrier, referring to an integrated hardware and software system that carries the data processing, model calculation, early warning triggering, and information push functions of the method of this invention.

[0018] The three-dimensional geological and tunnel model of the mine in this embodiment is constructed based on geological exploration data (boreholes, geophysical exploration), tunnel design drawings (CAD), and real-time monitoring data (such as the location of roof separation). It includes the stratigraphic structure (such as the thickness of coal seams and rock strata), tunnel topology (such as the layout of mining areas and chambers), the location of the mining face (such as the current advance), and key facilities (such as pump rooms and refuge chambers).

[0019] It should be noted that the technical means not described in detail in the following embodiments are all conventional means in the field, are not the key points of the invention, and will not be elaborated upon.

[0020] See Figure 1 This embodiment provides a method for real-time monitoring and intelligent early warning of underground mine disasters based on multi-sensor fusion, including the following steps: Step 1: Multi-source heterogeneous data acquisition and fusion preprocessing Through a sensor network deployed underground (including hydrological monitoring sensors, temperature sensors, gas monitoring sensors, dust concentration sensors, roof delamination sensors, and anchor bolt / cable stress sensors) and a remote sensing monitoring system deployed on the surface (including UAV remote sensing equipment and satellite remote sensing equipment), multi-source heterogeneous real-time monitoring data related to water hazards, fires, gas hazards, dust hazards, roof hazards, mine pressure hazards, and geological hazards are collected. The underground sensor network covers hazard types such as water hazards (hydrological monitoring), fires (temperature), gas (gas concentration), dust (dust concentration), roof hazards (delamination sensors), and mine pressure (anchor bolt / cable stress), directly sensing underground environmental parameters. The surface remote sensing monitoring system uses UAVs (visible light and infrared imagery) + satellites (synthetic aperture radar interferometry InSAR) to monitor geological hazards (surface subsidence, ground fissures). In this embodiment, the monitoring frequency of the UAV remote sensing equipment and the satellite remote sensing equipment is dynamically adjusted according to the surface deformation rate. When the surface displacement rate exceeds the preset threshold, the frequency of drone aerial photography is increased to once a day, and the frequency of satellite remote sensing data updates is increased to once every 12 hours; otherwise, the normal frequency is maintained, thereby optimizing the use of remote sensing resources (intensified monitoring during high-risk periods and cost savings during low-risk periods) and ensuring the timely detection of geological disasters (such as precursors to collapses). The collected multi-source heterogeneous real-time monitoring data undergoes fusion preprocessing, including timestamp alignment, spatial location registration, format standardization, and outlier cleaning, to generate a standardized data stream with spatiotemporal uniformity. This standardized data stream is then uploaded to the integrated management and control platform. By acquiring comprehensive disaster monitoring data, the issues of multi-source data format and spatiotemporal inconsistencies are resolved, providing standardized input for subsequent evaluation. The specific steps of the fusion preprocessing are as follows: Timestamp alignment: Unify the time base for data acquisition from different sensors (e.g., based on UTC time) to ensure timing consistency; Spatial location registration: Establish a three-dimensional mapping relationship between all sensor data and the unified geographic coordinate system of the mine, so that each data point is associated with its underground roadway location (such as X / Y / Z coordinates, based on underground GPS or laser scanning) or surface coordinates (such as latitude, longitude and elevation). This is achieved by a streaming data fusion algorithm based on a spatiotemporal hypergraph model. Specifically, a dynamic spatiotemporal hypergraph is constructed by taking each sensor node with different sampling rates, which may have transmission delays or data loss, and its observations within a sliding time window. When data flows in, the spatiotemporal coordinate confidence of each data point is iteratively calculated in the hypergraph through a confidence propagation algorithm. High-confidence nodes (such as fixed roof separation instruments) are used to correct the spatiotemporal information of low-confidence nodes (such as mobile device sensors), thus achieving soft registration of asynchronous heterogeneous data. Format standardization: Convert the proprietary protocols of different sensors (such as Modbus and CAN bus) into a unified data format (such as JSON and CSV). Outlier cleaning: Remove abrupt data caused by sensor malfunctions (e.g., using the 3σ criterion to identify outliers); After the above fusion preprocessing, the output standardized data stream is a series of feature tuples with confidence weights and association labels; for example, a feature tuple can be represented as: {Time window: T, Spatial domain: S, Feature set: {Gas concentration_anomaly, Temperature gradient_positive}, Correlation strength: α, Overall confidence: β} This data stream not only contains monitoring values, but also contains spatiotemporal correlation information between multiple data sources, providing structured input for subsequent coupled risk analysis; In this embodiment, when monitoring geological disaster-related data, visible light and infrared image data taken regularly by UAVs (to identify surface cracks) and synthetic aperture radar interferometry data provided by satellites (to monitor millimeter-level surface subsidence) are integrated. The displacement field information (such as subsidence rate and crack width) of surface collapse and ground cracks is quantitatively obtained through image processing (such as edge detection to extract cracks) and deformation inversion algorithms (such as PS-InSAR inversion displacement field), and is used as part of a standardized data stream. Step 2: Construction of a two-layer safety status assessment model and dynamic weight assignment Based on a combination of the Analytic Hierarchy Process (AHP) and expert experience, a two-layer safety status evaluation model is constructed within the integrated management and control platform. This model includes a lower layer of safety status evaluation models for several disaster subsystems and an upper layer of an overall mine safety status evaluation model. The disaster subsystem safety status evaluation models construct separate evaluation models for seven types of disasters, including water, fire, and gas (e.g., the gas subsystem model includes indicators such as gas concentration and emission rate). The overall mine safety status evaluation model integrates the scores of each subsystem to assess the overall mine risk (considering the coupling effects between subsystems). In this embodiment, a disaster coupling amplification coefficient is introduced when weighting the overall mine safety status evaluation model. This coefficient is quantified by constructing a dynamic Bayesian network across disaster subsystems, representing the key risks of different disasters. Nodes (such as gas concentration and water inflow) serve as nodes in a Bayesian network. Based on historical chain accident data, conditional probability tables between nodes are learned. When the risk of a certain subsystem increases, the state of its corresponding node changes. Through network inference, the prior probability of risk nodes in other related subsystems is dynamically affected, thereby achieving real-time and quantitative amplification of coupled risks. This is used to correct subsystem weights to quantitatively characterize the associated risks of different disaster types. Based on historical accident analysis (such as cases where gas explosions cause roof collapse), a disaster coupling amplification coefficient is defined (such as roof instability risk weight × 1.5 when gas is abnormal). This corrects subsystem weights to quantify associated risks, overcoming the limitations of traditional single-hazard assessment and reflecting the risk amplification effect of multi-hazard chain reactions (such as gas-fire coupling and roof-water permeability coupling). Using historical data and expert questionnaires, initial weights are assigned to the evaluation indicators in the safety status evaluation models of each disaster subsystem, and initial weights are assigned to the contribution of each subsystem in the overall mine safety status evaluation model. (The core of dynamic weight assignment lies in the Bayesian inference mechanism based on real-time evidence. The evaluation indicators of each disaster subsystem are modeled as a dynamic Bayesian network. When real-time monitoring data is input as evidence, the network infers using the sequential Monte Carlo method, not only outputting the risk probability but also inferring the instantaneous contribution of each indicator to the current overall risk status in the current specific context. This contribution is the dynamic weight of the indicator. This mechanism allows the weight to change in real time with environmental conditions (such as a sudden decrease in ventilation). The specific operation steps are as follows: a. A group of experts, based on their experience, compares the importance of each evaluation indicator within the same level pairwise to construct a judgment matrix; b. Perform consistency checks and corrections on the judgment matrix; c. The judgment matrix that has passed the test is weighted and integrated with the objective weighting factors derived from the statistical analysis of historical disaster data to form the final comprehensive weight; By balancing expert experience with data objectivity, and avoiding the biases of a single scoring method (such as the possibility that purely subjective methods might overlook high-frequency exceeding indicators), the following is achieved: Initial scoring: Determine the initial weights by combining historical disaster data (such as the frequency of gas exceedance accidents in a certain mine) and expert questionnaires (such as the scores given by 10 experts on the importance of "gas concentration"); Dynamic adjustment: Optimize weights based on feedback from subsequent early warning effects (step five) to adapt to environmental changes (such as an increase in mining depth leading to an increase in the weight of ore pressure). The weighted scoring results are set to be dynamically adjusted based on subsequent feedback. By constructing a quantitative evaluation model, the safety status of each disaster subsystem and the overall mine is dynamically assessed. Step 3: Real-time comprehensive assessment and early warning classification of disaster risks The standardized data stream obtained in step one is input in real time into the two-layer safety status evaluation model constructed in step two. First, the safety status evaluation models of each disaster subsystem are calculated in parallel, and the real-time safety status score and risk level of each disaster subsystem are output. Each subsystem model runs independently (e.g., the gas model only processes gas data) to improve efficiency. The real-time comprehensive evaluation is achieved by simulating the competition-cooperation relationship between disasters through a multi-agent reinforcement learning framework. Each disaster subsystem is regarded as an agent, whose goal is to accurately assess its own risk. All agents interact in a partially observable Markov game environment. The upper-level mine overall model integrates an attention mechanism network to dynamically observe the behavior (output score) of each agent and the environmental state, and learns to assign attention weights to the output of each agent under different global situations. The final risk level and score of the mine as a whole are the result generated by the attention network aggregating the information of each agent after the multi-agent system reaches game equilibrium, rather than a simple weighted average. Subsequently, the scores and risk level results of each subsystem are input into the overall mine safety status evaluation model to calculate the real-time comprehensive safety status score and risk level of the mine as a whole. By calculating the risk level of each subsystem and the overall mine in real time, a basis for early warning is provided. The risk level is divided into four levels, including safety, attention, early warning and danger. The threshold of each level is determined comprehensively based on the industry safety standards of each type of disaster, the mine's own geological conditions and historical data distribution characteristics. By unifying the risk classification standard, subjective arbitrariness is avoided and the consistency of early warning triggering is ensured. In this embodiment, the calculation of the safety status evaluation model of each disaster subsystem adopts the fuzzy comprehensive evaluation method, which transforms continuous monitoring data (such as gas concentration of 23ppm) into a discrete evaluation set (such as low, medium, and high), and outputs a safety status score by combining the weighted scoring results, so as to solve the fuzziness of monitoring data (such as the difficulty in accurately quantifying slightly high concentrations) and the compatibility of evaluation models, and improve the objectivity of subsystem scoring. Step 4: Matching and Pushing Intelligent Early Warning Measures The integrated management and control platform automatically triggers an early warning mechanism if the risk level of any disaster subsystem or the mine as a whole, as output in step three, reaches a preset threshold. Based on the specific disaster type and risk level triggered, the integrated management and control platform intelligently matches and generates structured early warning information from a pre-set expert knowledge rule base. This information includes specific handling suggestions (such as starting drainage pumps), responsible personnel (such as the head of ventilation), and execution procedures (such as arriving at the scene within 10 minutes). The information is then pushed in real time through pre-set multi-channels (SMS, APP, audible and visual alarms). The expert knowledge rule base has a pre-set mapping of "disaster type - risk level - handling measures" (such as "gas warning" corresponding to "stop operation and strengthen ventilation"). In this embodiment, the expert knowledge rule base is constructed using an ontology-based knowledge representation method. It formally defines and stores disaster entities (such as gas), monitoring indicators (such as concentration), risk status (such as early warning), response actions (such as ventilation), resource objects (such as ventilators) and their interrelationships (such as "gas concentration exceeds the standard → trigger early warning → execute ventilation"), forming a reasonable knowledge network that supports automatic rule matching by computers. The specific process of intelligent matching and generating structured early warning information is as follows: Based on a knowledge network, graph traversal (e.g., starting from a gas early warning node to find related response actions) and rule-based reasoning (e.g., if the gas concentration is >1% and the wind speed is <0.5m / s, then the backup fan should be activated first), multiple basic response rules are automatically combined (e.g., notifying personnel to evacuate + activating the fan + reporting to the dispatch room) to form an operable comprehensive emergency plan adapted to specific complex scenarios. This plan can handle complex situations (e.g., multiple indicators exceeding the standard simultaneously), generating operable combined measures rather than single response suggestions. The core of intelligent matching is plan generation based on causal reasoning. The method involves constructing an expert knowledge rule base that is a time-varying causal knowledge graph. Nodes within this graph have explicit causal edges such as 'cause' and 'inhibit'. When an alert is triggered, the system first performs causal discovery on the current multidimensional feature tuple flow to pinpoint the key causal path leading to the risk state. Subsequently, it performs counterfactual reasoning to simulate the impact of different intervention measures (such as 'increasing ventilation') on the risk chain. The resulting contingency plan is an executable and interpretable dynamic action plan that integrates causal explanations and multi-scenario deduction results (such as cost-effectiveness estimates of different measures), rather than a simple query result of static rules. Step 5: Early Warning Effect Feedback and Model Adaptive Optimization The integrated management and control platform records complete data for each early warning event (such as gas concentration curve), the implementation status of the pushed measures (such as whether ventilation was started), and the final treatment effect (such as whether the concentration decreased after 30 minutes). Based on records, the weighting parameters in step two are periodically optimized using reinforcement learning algorithms to enable the overall mine safety status evaluation model to adapt to changes in the mining stage and geological conditions. The model is iteratively optimized using historical early warning data to improve accuracy. The iterative optimization using reinforcement learning algorithms specifically refers to the use of offline reinforcement learning and counterfactual evaluation methods to ensure the safety and efficiency of the optimization process. Specifically, the system constructs an offline safety decision dataset from all historical early warning decisions, response actions, and subsequent results. During optimization, offline strategy evaluation methods such as double robust estimation are used to evaluate the long-term safety benefits that the new model weights may bring without actually implementing the new strategy. The model update adopts the conservative Q-learning algorithm, whose optimization objective is to maximize the expected safety benefits while strictly constraining the deviation between the new strategy and the historically verified safety strategies, thereby avoiding the exploration of unknown high-risk strategies and achieving adaptive optimization of the model under safety constraints. Step Six: 3D Visualization and Situation Simulation The pre-built 3D geological and tunnel model of the mine integrated into the comprehensive management and control platform dynamically displays the real-time monitoring data in step one, the risk level heat map output in step three, and the early warning location and information triggered in step four; and performs short-term simulation and deduction of disaster development trends based on current data, and assists decision-making by intuitively presenting risk distribution and trends.

[0021] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for real-time monitoring and intelligent early warning of underground mine disasters based on multi-sensor fusion, characterized in that, Includes the following steps: Step 1: Collect multi-source heterogeneous real-time monitoring data related to underground mine disasters through underground sensor networks and surface remote sensing monitoring systems, perform data fusion preprocessing, generate a standardized data stream with unified spatiotemporal characteristics, and upload it to the integrated management and control platform; Step 2: Combining the Analytic Hierarchy Process (AHP) with expert experience, a two-layer safety status evaluation model is constructed within the integrated management and control platform, consisting of a safety status evaluation model for several disaster subsystems at the lower level and a safety status evaluation model for the overall mine at the upper level. Historical data and expert questionnaires are used to dynamically assign weights. Step 3: Input the standardized data stream into the two-layer safety status evaluation model in real time. The safety status evaluation models of each disaster subsystem calculate and output scores and risk levels in parallel, and input them into the overall mine safety status evaluation model to obtain the overall real-time comprehensive safety status score and risk level of the mine. Step 4: When the risk level reaches the threshold, the integrated management and control platform triggers an early warning, intelligently matching and generating structured information containing disposal suggestions, responsible personnel, and execution procedures from the expert rule base according to the disaster type and level, and pushing it through multiple channels; Step 5: The integrated management and control platform records the data of early warning events, the implementation of measures and the effects of handling them, and regularly uses reinforcement learning to iteratively optimize the dynamic weighting parameters in Step 2, so that the overall safety status evaluation model of the mine can adapt to changes in mining stages and geological conditions. Step Six: In the integrated 3D geological and roadway model of the mine integrated management and control platform, dynamically display the multi-source heterogeneous real-time monitoring data collected in Step One, the heat map of risk level output in Step Three, the early warning information triggered in Step Four, and short-term simulation of disaster trends.

2. The method for real-time monitoring and intelligent early warning of underground mine disasters based on multi-sensor fusion as described in claim 1, characterized in that: In step one, the downhole sensor network includes hydrological monitoring sensors, temperature sensors, gas monitoring sensors, dust concentration sensors, roof delamination meters, and anchor bolt and cable stress sensors. The surface remote sensing monitoring system includes UAV remote sensing equipment and satellite remote sensing equipment. The multi-source heterogeneous real-time monitoring data includes data related to water hazards, fires, gas disasters, dust disasters, roof disasters, mine pressure disasters, and geological disasters. The data fusion preprocessing includes timestamp alignment, spatial location registration, format standardization, and outlier cleaning.

3. The method for real-time monitoring and intelligent early warning of underground mine disasters based on multi-sensor fusion according to claim 2, characterized in that: The spatial registration establishes a three-dimensional mapping relationship between all sensor data and the unified geographic coordinate system of the mine, so that each data point is associated with the location of the underground roadway or the surface coordinates. The geological disaster-related data monitoring integrates UAV visible light and infrared image data and satellite synthetic aperture radar interferometry data. After image processing and deformation inversion, the displacement field information of surface collapse and ground fissure is quantitatively obtained as part of the standardized data stream.

4. The method for real-time monitoring and intelligent early warning of underground mine disasters based on multi-sensor fusion as described in claim 1, characterized in that: In step two, the specific steps for dynamically assigning weights are as follows: a) A group of experts constructs a judgment matrix by comparing each pair of evaluation indicators at the same level; b. Examine and correct the judgment matrix; c. Integrate the judgment matrix that has passed the test with the objective weighting factors of historical data statistics to form a comprehensive weight.

5. The method for real-time monitoring and intelligent early warning of underground mine disasters based on multi-sensor fusion according to claim 1, characterized in that: In step two, a disaster coupling amplification coefficient is introduced to weight the overall mine evaluation model. The disaster coupling amplification coefficient is derived from the analysis of the chain co-occurrence and aggravation relationship of multiple disasters in historical accidents, and is used to correct the subsystem weights to quantify the disaster-related risks.

6. The method for real-time monitoring and intelligent early warning of underground mine disasters based on multi-sensor fusion according to claim 1, characterized in that: In step three, the risk level is divided into four levels: safe, caution, warning, and danger. The threshold is determined comprehensively based on industry standards, mine geological conditions, and historical data distribution.

7. The method for real-time monitoring and intelligent early warning of underground mine disasters based on multi-sensor fusion according to claim 1, characterized in that: In step four, the expert rule base uses ontology knowledge representation to formally define and store disaster entities, monitoring indicators, risk status, response actions, resource objects and relationships, forming a reasonable knowledge network.

8. The method for real-time monitoring and intelligent early warning of underground mine disasters based on multi-sensor fusion according to claim 1, characterized in that: In step four, intelligent matching generates structured information by performing graph traversal and rule-based reasoning based on a reasonable knowledge network, and combining basic handling rules to form an operable comprehensive plan that adapts to complex scenarios.

Images