Coal mine safety assessment method and system based on large model enabling

Abstract

The invention relates to the technical field of coal mine safety assessment, in particular to a coal mine safety assessment method and system based on large model energization, and the method comprises the following steps: determining a plurality of coal mine disaster types and a plurality of joint common parameters, and building an association relationship between the plurality of joint common parameters and the plurality of coal mine disaster types; determining abnormal joint common parameters as an abnormal common parameter group based on a preset condition, generating an early warning label based on the abnormal common parameter group, and determining a coal mine disaster type corresponding to the abnormal common parameter group based on the association relationship to obtain at least one candidate disaster type; and inputting the early warning label into a pre-constructed early warning model to obtain an identification result containing disaster risk probability distribution corresponding to each candidate disaster type, and when the risk probability of the disaster corresponding to any candidate disaster type exceeds a corresponding preset threshold, automatically triggering an early warning signal, so that the safety risk can be identified in advance, and early warning can be performed.

Description

Technical Field

[0001] This invention relates to the field of coal mine safety assessment technology, specifically to a coal mine safety assessment method and system based on large-scale model empowerment. Background Technology

[0002] Coal mine safety production faces multiple hazards, including water, fire, gas, coal dust, and roof collapse. Traditional coal mine safety assessments typically employ fixed threshold alarms, triggering an alarm when a monitored parameter exceeds a preset threshold. However, coal mine disasters are often the result of coordinated anomalies across multiple parameters, and single-parameter alarms suffer from high false alarm rates and delayed warning times. Current technologies lack the comprehensive analytical capabilities for multi-parameter correlation anomalies. When multiple monitored parameters have not reached their individual alarm thresholds but exhibit a coordinated anomaly trend, it is difficult to accurately identify potential disaster types in the early stages of a disaster, affecting the timeliness and accuracy of coal mine safety assessments.

[0003] To address these issues, we propose a coal mine safety assessment method and system based on a large model. Summary of the Invention

[0004] The purpose of this invention is to provide a coal mine safety assessment method and system based on large model empowerment to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a coal mine safety assessment method and system based on a large model, the method comprising the following steps: Identify multiple coal mine disaster types and multiple joint common parameters, and establish the correlation between multiple joint common parameters and multiple coal mine disaster types; Based on preset conditions, joint common parameters that are abnormal are identified as abnormal common parameter groups. Early warning labels are generated based on abnormal common parameter groups. Based on the correlation, the coal mine disaster type corresponding to the abnormal common parameter group is determined to obtain at least one candidate disaster type. The warning label is input into the pre-built warning model to obtain the identification result containing the disaster risk probability distribution corresponding to each candidate disaster type. When the risk probability of any candidate disaster type exceeds the corresponding preset threshold, the warning signal is automatically triggered. Based on the coal mine disaster type indicated by the early warning signal, a corresponding three-dimensional disaster evolution scenario is generated. In the generated disaster evolution scenario, an emergency rescue drill or safety training module is configured. By receiving and responding to external user instructions, the module drives the three-dimensional scenario to change in response to user operations, executes interactive drills or training processes according to user instructions, and outputs drill results or training feedback.

[0006] Preferably, the step of determining multiple coal mine disaster types and multiple joint common parameters includes: Identify multiple types of coal mine disasters and configure a joint common parameter cluster for each type of coal mine disaster; Multiple joint common parameter clusters were obtained based on various coal mine disaster types; Obtain all joint common parameters from multiple joint common parameter groups, and extract all non-repeating joint common parameters to obtain multiple joint common parameters.

[0007] Preferably, the step of establishing the correlation between multiple joint common parameters and multiple coal mine disaster types includes: Multiple coal mine disaster types and several common parameters are obtained and stored in a type database and a parameter database, respectively. The type database is divided into multiple subtype databases, each of which corresponds to a type of coal mine disaster. Multiple nodes are set up in the parameter database and connected to form a node network, where each node corresponds to a joint common parameter; Configure a data chain for the node, the data chain containing multiple data points arranged in a time series; Establish a sensing chain between the subtype library and the node network to form a correlation between coal mine disaster types and joint common parameters.

[0008] Preferably, the step of establishing the inductive link between the subtype library and the node network includes: Multiple information points are configured for each subtype library, and the parameters in the joint common parameter group corresponding to each subtype library are distributed and stored in the information points. Configure a corresponding sensing point for each information point, and establish a sensing chain between sensing points and nodes that store the same joint common parameters.

[0009] Preferably, the step of determining the joint common parameters that are abnormal based on preset conditions as an abnormal common parameter group, and generating a warning label based on the abnormal common parameter group includes: Configure preset conditions for each joint common parameter, collect real-time monitoring data of the joint common parameter in the coal mine, and take the joint common parameter corresponding to the real-time monitoring data that does not meet the corresponding preset conditions as the abnormal common parameter. Based on the synchronicity of parametric anomaly time series, multiple common anomaly parameters are combined to obtain anomaly common parameter clusters; Extract the feature information of the abnormal common parameter clusters and generate corresponding early warning labels.

[0010] Preferably, the step of determining the coal mine disaster type corresponding to the abnormal common parameter cluster based on the correlation relationship to obtain at least one candidate disaster type includes: Obtain multiple common abnormal parameters and corresponding abnormal nodes from at least one abnormal common parameter group, and activate the sensing points of the corresponding abnormal nodes in each subtype library based on the sensing chain. In the subtype library, the information points corresponding to the sensing points are differentiated to obtain the marked information points. Among them, the information points of the same abnormal node in different subtype libraries are marked in the same way, and the information points corresponding to different abnormal nodes are marked in different ways. The disaster type corresponding to the subtype library with marked information points is used as the candidate disaster type.

[0011] Preferably, the step of inputting the warning label into a pre-constructed warning model to obtain the disaster risk probability distribution corresponding to each candidate disaster type, and automatically triggering a warning signal when the probability of any disaster risk exceeds the corresponding preset threshold, includes: Input the warning labels into the pre-built warning model; The early warning model is used to calculate the identification results, which include the probability distribution of disaster risk corresponding to each candidate disaster type. Monitor the disaster risk probability corresponding to each candidate disaster type, and automatically trigger the corresponding early warning signal when the disaster risk probability of any candidate disaster type exceeds its corresponding preset threshold.

[0012] A coal mine safety assessment system based on large model empowerment, applied to any one of the above-mentioned coal mine safety assessment methods based on large model empowerment, comprising: The relationship building module is used to determine multiple coal mine disaster types and multiple joint common parameters, and to establish the correlation between multiple joint common parameters and multiple coal mine disaster types; The determination module is used to determine the joint common parameters that are abnormal based on preset conditions as an abnormal common parameter group, generate early warning labels based on the abnormal common parameter group, and determine the coal mine disaster type corresponding to the abnormal common parameter group based on the correlation to obtain at least one candidate disaster type; The risk assessment module is used to input the warning labels into the pre-built warning model to obtain the identification results containing the disaster risk probability distribution corresponding to each candidate disaster type. When the risk probability of any candidate disaster type exceeds the corresponding preset threshold, the warning signal is automatically triggered. The scenario simulation module is used to generate a corresponding three-dimensional scenario of disaster evolution based on the coal mine disaster type indicated by the early warning signal. In the generated disaster evolution scenario, an emergency rescue drill or safety training module is configured. By receiving and responding to external user instructions, the module drives the three-dimensional scenario to change in response to user operations, executes interactive drills or training processes according to user instructions, and outputs drill results or training feedback.

[0013] Compared with the prior art, the beneficial effects of the present invention are: By using sensing chains and differentiated markers, anomalies in low-level parameters are automatically and in real time mapped to high-level disaster type hypotheses, enabling automatic identification of disaster risks. Instead of viewing individual parameter anomalies in isolation, the system focuses on the matching degree between combinations of multiple anomaly parameters and correlation patterns of different disaster types. This allows for more accurate determination of disaster types. Even if some parameters do not reach fixed alarm thresholds, as long as multiple parameters show a coordinated abnormal trend and can activate correlation patterns of specific disaster types, risks can be identified in advance, providing early warnings and improving the response speed and accuracy of coal mine safety assessment and management. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram illustrating the relationships within the present invention; Figure 3 This is a system structure block diagram 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] Example: Please refer to Figures 1 to 3 This invention provides a coal mine safety assessment method and system technical solution based on a large model: A coal mine safety assessment method based on a large model includes the following steps: S1: Identify multiple coal mine disaster types and multiple joint common parameters, and establish the correlation between multiple joint common parameters and multiple coal mine disaster types; The steps for determining multiple coal mine disaster types and multiple joint common parameters include: determining multiple coal mine disaster types, configuring a joint common parameter group for each coal mine disaster type; obtaining multiple joint common parameter groups based on multiple coal mine disaster types, obtaining all joint common parameters in the multiple joint common parameter groups, and extracting all non-repeating joint common parameters to obtain multiple joint common parameters; Specifically, multiple joint common parameter clusters are obtained, each corresponding to a type of coal mine disaster; a union operation is performed on each joint common parameter cluster to obtain all joint common parameters; deduplication is performed on all joint common parameters to extract all unique joint common parameters; the union operation uses set operation methods to merge all parameters in each parameter cluster into a single parameter set; the deduplication process includes: identifying identical parameters in different parameter clusters; retaining unique parameter identifiers; and recording the frequency of occurrence of the parameter in all parameter clusters.

[0018] Suppose there are three joint common parameter clusters: parameter cluster 1 (flood): {humidity, roof pressure, temperature}, parameter cluster 2 (gas accident): {gas concentration, temperature, wind speed}, and parameter cluster 3 (coal dust accident): {coal dust concentration, temperature, wind speed}. After performing a union operation and removing duplicates, multiple unique joint common parameters are obtained. These multiple unique joint common parameters include humidity, roof pressure, temperature, gas concentration, wind speed, and coal dust concentration, which are used to establish the correlation between multiple joint common parameters and disaster types.

[0019] Multiple coal mine disaster types are identified, including flooding, gas accidents, coal dust accidents, coal seam spontaneous combustion accidents, external fire accidents, and roof collapse accidents. The monitoring parameter set configured for flooding includes at least one of temperature, humidity, roof pressure, gas concentration, and carbon monoxide concentration. The monitoring parameter set configured for gas accidents includes at least one of methane concentration, oxygen concentration, temperature, and wind speed. The monitoring parameter set configured for coal dust accidents includes at least one of coal dust concentration, equipment operating temperature, and date information. For each disaster type, a set of joint common parameters for early identification is configured as a joint common parameter group for that disaster type. The joint common parameter group consists of multiple joint common parameters, including temperature, humidity, gas concentration, and carbon monoxide concentration. At least one of the following: temperature, methane concentration, oxygen concentration, roof pressure, coal dust concentration, and equipment operating temperature. Joint common parameters refer to measurable physical or chemical quantities that can predict the impending occurrence of one or more specific disasters through abnormal changes before a coal mine disaster. Joint common parameters are monitoring parameters that reflect multiple disaster precursors. These parameters must appear in the precursor characteristics of at least two disaster types, and the monitoring thresholds differ across different disaster types. All these joint common parameters are continuously monitored. When a joint common parameter (such as temperature) changes, an alarm is not immediately triggered. Instead, other related parameters are retrieved to see if they also show the expected coordinated changes. The observed coordinated change patterns of this set of parameters are packaged into a warning label. For example, a warning label is generated: "Pattern_Flood_Suspected": {Temperature decrease, humidity increase, pressure fluctuation}. This warning label is then fed into the warning model. The model calculates the probability of a flood actually occurring when it observes the label 'Pattern_Flood_Suspected'. It doesn't look at individual values ​​exceeding a fixed threshold, but rather uses an early warning model to predict the probability during the "incubation period" of a disaster based on the appearance of these common abnormal parameter clusters, improving the analysis and assessment of coal mine safety. It integrates and deduplicates the joint common parameter clusters of all disaster types and establishes the correlation between joint common parameters and disaster types. The parameter "temperature" is a warning indicator for gas explosions, fires, coal dust explosions, and coal seam spontaneous combustion. Therefore, it is established as a joint common parameter. By monitoring one parameter, such as temperature, it can simultaneously participate in the risk assessment of multiple disasters (such as gas, fire, coal dust, and spontaneous combustion), greatly improving the efficiency of the monitoring system. When temperature is abnormal, it doesn't judge in isolation, but rather checks whether other related joint common parameters (such as gas concentration and coal dust concentration) are also abnormal, thus comprehensively judging the most likely disaster type and achieving more accurate early warning.

[0020] The steps for establishing the correlation between multiple joint common parameters and multiple coal mine disaster types include: acquiring multiple coal mine disaster types and multiple joint common parameters, and storing them in a type database and a parameter database, respectively; dividing the type database into multiple sub-type databases, where each sub-type database corresponds to one coal mine disaster type; setting multiple nodes in the parameter database and connecting them to form a node network, where each node corresponds to one joint common parameter; configuring data chains for the nodes, where the data chains contain multiple data points arranged in a time series; and establishing a sensing chain between the sub-type databases and the node network to form the correlation between coal mine disaster types and joint common parameters.

[0021] Multiple coal mine disaster types and several common parameters are obtained and stored in a type database and a parameter database, respectively. Both the type database and the parameter database are configured in a cloud server. Specifically, the raw data from business systems such as gas, water hazards, and rock bursts at the corresponding collection points are jointly stored in an integrated parameter database. The parameter database stores multi-source data, such as personnel safety data, environmental data, hydrological data, and fire data (data related to assessing coal mine risks, which may correspond to different risks, such as personnel safety risks or fire risks, etc.). It should be noted that the data points in the data chain are synchronized based on a unified timestamp, supporting the simultaneous acquisition of numerical sequences of multiple joint common parameters within the same time period. A data chain containing multiple data points is configured for each node, and the data points store the values ​​of the joint common parameters in a time sequence. A unified timestamp is configured for all data points for time synchronization. A timestamp index is established to support the simultaneous acquisition of numerical sequences of multiple joint common parameters within the same time period. Simultaneously acquiring multiple parameter numerical sequences includes: receiving a time range query request; locating all data points within the corresponding time period based on the timestamp index; and returning the synchronized numerical sequence of multiple parameters within that time period.

[0022] The steps for establishing an induction link between a subtype library and a node network include: configuring multiple information points for the subtype library, distributing the parameters in the joint common parameter group corresponding to each subtype library in the information points; configuring a corresponding induction point for each information point, and establishing an induction link between the induction points storing the same joint common parameters and the nodes.

[0023] It should be noted that the parameter database is responsible for managing, storing, and circulating joint common parameters (such as raw data of temperature, gas concentration, etc.), and is a central repository storing the original definitions of all joint common parameters. A node is a logical unit in the parameter database, uniquely corresponding to a joint common parameter (for example, one node is specifically responsible for "methane concentration," and another node is specifically responsible for "temperature"). A node is the "representative" or processor of the joint common parameter in the logical network, responsible for receiving, temporarily storing, and processing the data of its corresponding parameter. The node network is a network topology structure formed by all nodes interconnected through communication channels, realizing the logical association between parameters. When different parameter nodes are activated (i.e., parameter anomalies), the node network can quickly identify this association, facilitating subsequent disaster type determination. Multiple nodes are set in the parameter database and connected through communication channels to form a node network. Trigger conditions are set for each communication channel (the trigger condition is that the joint common parameters corresponding to both ends of the communication channel are both abnormal common parameters). The communication channel is a logical path connecting two nodes, specifying which parameters are related. Its trigger condition (both ends of the parameters are abnormal) is the key mechanism for initiating intelligent early warning, ensuring the combination of multiple abnormal common parameters; the data chain is a time-ordered data structure attached to the nodes, specifically used to store the historical and real-time numerical sequences of the parameters corresponding to the node, recording the changes of the parameters over time. A data point is the smallest storage unit in the data chain, recording the specific value of the parameter corresponding to its node at a specific moment. The format is usually (timestamp, parameter value); the type database is mainly responsible for managing and representing the knowledge of coal mine disaster types (such as floods, gas accidents, etc.), and is used to store the central repository of all coal mine disaster type definitions; the subtype library is a sub-unit in the type database, uniquely corresponding to a coal mine disaster type (e.g., the gas accident subtype library). It is a knowledge base for a certain disaster type, containing all the joint common parameters required to determine the disaster type; information points are distributed in multiple, communicable storage units in the subtype library, distributing the joint common parameter groups required for a subtype library (a disaster). For example, a gas accident subtype library might have three information points, storing the definitions and statuses of parameters such as "humidity," "roof pressure," and "temperature," respectively. A sensing point, configured as a proxy or interface on an information point, can sense or monitor anomalies in the common parameters of the nodes corresponding to its information point. The sensing chain is a bidirectional data channel connecting sensing points and nodes, establishing the correlation between "parameters" and "disaster types." When the data of a parameter node becomes abnormal, the anomaly signal can be instantly transmitted through the sensing chain to all disaster subtype libraries that are interested in that parameter; conversely, the subtype libraries can also proactively obtain the latest data for the required parameters through the sensing chain.

[0024] Taking the anomalies of two parameters, "methane concentration" and "temperature," as an example: Real-time data of "methane concentration" and "temperature" are collected by sensors. The "methane concentration node" and "temperature node" in the node network each receive the data and determine it as an anomaly. A new data point is added to their respective data chains. Because there is a communication channel between the "methane concentration node" and the "temperature node," and the triggering condition is met (both ends are anomaly), this channel is activated, indicating the emergence of a meaningful anomalous parameter cluster {methane concentration, temperature}. The anomaly signal is instantly transmitted from the "methane concentration node" and "temperature node" to all the sensor points connected to them through the sensor chain. The information points of these sensor points receive the signal. The subtype library to which these information points belong (e.g., the gas accident subtype library) is activated because its internal information points exactly store the parameter cluster {methane concentration, temperature}. Thus, "gas accident" is determined as a candidate disaster type. The gas accident subtype library requests historical data sequences with a unified timestamp from nodes through the sensing chain. By separating common parameters and disaster types through the node network and sensing chain, and then dynamically linking them through flexible connections, it achieves accurate identification and candidate screening of coal mine disaster types.

[0025] S2: Based on preset conditions, determine the joint common parameters that are abnormal as the abnormal common parameter group, generate early warning labels based on the abnormal common parameter group, and determine the coal mine disaster type corresponding to the abnormal common parameter group based on the correlation to obtain at least one candidate disaster type; The steps for determining abnormal joint common parameters as abnormal common parameter groups based on preset conditions and generating early warning labels based on these abnormal common parameter groups include: configuring preset conditions for each joint common parameter, the preset conditions including at least one of the following: the parameter value exceeds a corresponding preset threshold, the parameter change rate exceeds a corresponding preset threshold, and the difference between the parameter value and historical data of the same period exceeds a corresponding preset threshold; collecting real-time monitoring data of the joint common parameters in underground coal mines, and taking the joint common parameters corresponding to the real-time monitoring data that do not meet the corresponding preset conditions as abnormal common parameters; combining multiple abnormal common parameters based on the synchronicity of the parameter abnormality time series to obtain an abnormal common parameter group; extracting the feature information of the abnormal common parameter group and generating a corresponding early warning label; the early warning label includes the abnormal common parameter group identifier, parameter abnormality details, abnormality start time and duration, and associated potential disaster type.

[0026] It should be noted that a data chain is configured for each node. This data chain contains multiple data points arranged in a time series, used to store historical values ​​of joint common parameters in a time series. This not only determines whether there is an anomaly at the current moment but also analyzes the changing trends of the parameters (such as accelerated rises, sustained high-level oscillations, etc.). Anomaly common parameter clusters are determined based on synchronicity. To determine whether anomalies of multiple joint common parameters are synchronized in time, precise, timestamped historical data sequences are required. The data chain provides this foundation, enabling the calculation of key indicators such as the time difference of anomaly start time and the overlap of durations. Data points are synchronized based on a unified timestamp, allowing interpolation and alignment of data for different parameters. This ensures that when analyzing synchronicity, data from the same moment is compared, reducing misjudgments and supporting collaborative analysis of multi-parameter time-series data. By checking the persistence of anomalies in the data chain (persistence refers to the number of consecutively anomalous parameter values ​​exceeding a preset number), it ensures that the anomalies combined into the parameter cluster are real and stable signals. Accurate correlation information can be determined from the behavior of parameters changing over time, rather than just based on isolated instantaneous data points, thereby improving the accuracy of early warnings.

[0027] The steps for combining multiple common abnormal parameters to obtain a common abnormal parameter cluster based on the synchronicity of abnormal time series parameters are as follows: Calculate the time difference of the start time of each abnormal parameter; when the time difference is less than a preset time window, it is determined that the parameter abnormalities are synchronous; combine the synchronous abnormal parameters into the same common abnormal parameter cluster. The preset time window (T_window) is calculated as follows: T_window = max( N * T_sample, T_propagation ), where T_sample is the data acquisition period, N is an integer greater than or equal to 2, and T_propagation is the statistical delay time of abnormal propagation determined according to system characteristics; the integer N ranges from 2 to 5. The common abnormal parameter cluster is formed using graph theory: each abnormal parameter is regarded as a node in the graph. If two abnormal parameters are synchronous, an edge is established between the corresponding two nodes. Finally, each common abnormal parameter cluster is determined by finding a connected subgraph in the graph. Specifically, extracting the feature information of the abnormal common parameter cluster includes: extracting the numerical features of the abnormal parameters, which include at least one of the current value, average value, maximum value, and trend of change; extracting the temporal features of the abnormal parameters, which include at least one of the abnormal start time, duration, and cycle of change; extracting the spatial features of the abnormal parameters, which include at least one of the monitoring point location and spatial distribution pattern; for each abnormal parameter in the parameter cluster, extracting its current monitoring value, average value and maximum value within a specific time window, and analyzing its trend of change (e.g., continuous increase, rapid decrease, stable fluctuation, etc.); recording the abnormal start time of the entire parameter cluster, calculating the total duration to date, and analyzing its cycle of change (if any); recording the monitoring point location information corresponding to all abnormal parameters, and analyzing the spatial distribution pattern of these monitoring points underground (e.g., concentrated in the coal mining face, distributed along the return airway, etc.); and constructing a structured early warning label based on the extracted feature information. The warning label includes the following fields: Common Anomaly Parameter Group Identifier: A unique ID to identify this abnormal event; Parameter Anomaly Details: A structured list detailing the name of each abnormal parameter within the parameter group, its current value, the anomaly level (e.g., mild, moderate, severe) according to preset rules, and the trajectory of its value over time; Anomaly Start Time and Duration: Clearly recording the exact start time and duration of this abnormal event; Associated Potential Disaster Types: Automatically matching one or more coal mine disaster types (e.g., gas accidents, fires, floods) that the current common anomaly parameter group may indicate based on preset associations. The warning label also includes derived information: Environmental context data, including working face location, geological conditions, and mining technology; Risk assessment results, including risk level, impact range, and urgency; and suggested response plans, including emergency measures, key inspection points, and response procedures. By integrating scattered abnormal parameters into a structured warning label, the standardization and enrichment of warning information are achieved. This facilitates subsequent processing by the warning model, enabling faster and more accurate responses to potential risks.

[0028] The steps for determining the coal mine disaster type corresponding to the abnormal common parameter cluster based on the correlation relationship to obtain at least one candidate disaster type include: obtaining multiple abnormal common parameters and corresponding abnormal nodes in at least one abnormal common parameter cluster; activating the sensing points of the corresponding abnormal nodes in each subtype library based on the sensing chain; differentially marking the information points corresponding to the sensing points in the subtype library to obtain marked information points, wherein the information points of the same abnormal node in different subtype libraries use the same marking method, and the information points corresponding to different abnormal nodes use different marking methods; and taking the disaster type corresponding to the subtype library with marked information points as the candidate disaster type. It may also include: sorting candidate disaster types according to the number of marked information points in each subtype library; based on the sorting result, inputting the warning labels into a pre-built warning model in priority order; the warning model calculates the risk probability of each candidate disaster type in turn, which allows the warning model to prioritize the most likely disasters; when the calculated probability of a certain disaster exceeds the threshold, the warning can be triggered in advance; the computing resources are concentrated on the most dangerous target, improving the response speed and reducing the operating load. The specific steps for calculating risk probability using the early warning model are as follows: Collect a large amount of historical coal mine safety monitoring data and corresponding confirmed disaster event records. Each training sample contains two parts: first, multi-parameter time-series data within a window before the disaster occurs (which can be extracted as features similar to "early warning labels"); second, the disaster type label ultimately resulting from the sample (such as gas accident, flood, etc.). Based on historical data, generate "abnormal common parameter clusters" and extract their features to form a structured training feature set. Features include, but are not limited to: abnormal parameter combination type, abnormal amplitude of each parameter, abnormal duration, and temporal synchronization relationship between parameters. Using the above feature set and disaster type labels, train a classification model that can output probabilities. This model learns the probability of various disaster types occurring given a set of specific parameter abnormal features (i.e., early warning labels). When the system generates a new early warning label, it is first converted into a feature vector X according to the same rules as in the training phase. The feature vector X is then input into the trained early warning model. The model performs internal calculations and outputs a probability distribution vector P = [p1, p2, ..., pn], where n is the total number of disaster types defined by the system. The current warning label indicates the predicted probability of disaster type i occurring, and satisfies the following conditions: This probability This refers to the "probability of disaster risk"; Determining the Threshold Range and Process: By retrospectively analyzing historical data, an empirical baseline threshold range can be determined. For example, for a highly hazardous "gas accident," the baseline probability threshold for triggering a warning might be set at 0.7 (70%); for a relatively slow-developing "roof deformation," the baseline threshold might be set at 0.85. This range is typically between 0.6 and 0.95, and the specific value needs to be adjusted based on the mine's historical data and acceptable warning sensitivity. For example: the baseline threshold for gas accidents The correction factor is 0.75. The current system determines that the working face is in a "fault-boundary period," indicating a high environmental risk level. The value is 0.9. Therefore, the currently effective dynamic threshold is... When the early warning model calculates the probability of a gas accident under the current conditions... When the value is 0.70 > 0.675, the system will automatically trigger a gas accident warning signal.

[0029] The specific steps for sorting candidate disaster types based on the number of marked information points in each subtype library are as follows: 1) Count the number of information points corresponding to different marking methods in each subtype library; 2) Calculate the ratio of the total number of marked information points to the total number of information points in each subtype library; 3) Prioritize candidate disaster types based on the ratio; Counting the number of information points for different marking methods in each subtype library includes: identifying all marked information points in the subtype library; 4) Classifying and counting by marking method type; 5) Recording the number and specific location of information points corresponding to each marking method; 6) Calculating the ratio of the total number of marked information points to the total number of information points includes: counting the total number of marked information points in the subtype library; 7) Obtaining the total number of information point configurations in the subtype library; 8) Calculating the percentage of marked information points: (Total number of marked information points / Total number of information points) × 10 0%; Prioritization based on proportion includes: sorting candidate disaster types from high to low according to the proportion of marked information points; for candidate disaster types with the same proportion, secondary sorting is performed according to the importance of the marking method; the final sort is performed by combining the spatial distribution characteristics of the marked information points. The more marked information points a subtype has, the higher the probability that its corresponding disaster type is considered the current major risk, and the higher its ranking. When sorting candidate disaster types, if the proportion of marked information points is the same, secondary sorting can be performed according to the importance of the marking method. The importance of the marking method is determined based on the following factors: the severity of the abnormality of the parameters corresponding to the marking method and the weight of the parameters corresponding to the marking method in disaster identification. Through proportion calculation and multi-level sorting, the accuracy of disaster type priority is ensured. It should be noted that the differentiated marking includes: assigning a unique visual identifier to each abnormal node; marking all associated information points of the same abnormal node in different subtype libraries with the same visual identifier; using different visual identifiers to distinguish information points corresponding to different abnormal nodes; the visual identifier includes at least one of the following distinguishing dimensions: color coding, graphic shape, texture pattern, and animation effect. By constructing a sensing chain and marking mechanism, intelligent identification and sorting of candidate disaster types can be achieved, enabling early identification of multi-parameter collaborative anomalies, improving the timeliness of early warning, accurately identifying the most important disaster risks through quantitative sorting, and automatically and in real time mapping the underlying parameter anomalies to high-level disaster type hypotheses through the "sensing chain" and differentiated marking, realizing the automatic identification of disaster risks. It does not view a single parameter anomaly in isolation, but focuses on the matching degree between the "cluster" composed of multiple parameter anomalies and the association patterns of different disaster types, thereby more accurately judging the disaster type. Even if some parameters do not reach the fixed alarm threshold, as long as multiple parameters show a collaborative anomaly trend and can activate the association pattern of a specific disaster type, risks can be identified in advance, early warnings can be issued, and the response speed and accuracy of coal mine safety management can be improved. Specifically, multiple abnormal common parameters and their corresponding nodes are obtained from the abnormal common parameter group and marked as abnormal nodes. The abnormal nodes send parameter abnormality notifications to the corresponding sensing points through the sensing chain and mark the information points in the corresponding subtype library. The information points corresponding to multiple abnormal nodes are marked differently to obtain marked information points. Each node is assigned a unique identifier to represent the identity information of the corresponding joint common parameter. The sensing points associated with the same node use the same identifier. The disaster types corresponding to the subtype library with marked information points are used as candidate disaster types. The candidate disaster types are sorted according to the number of marked information points. The multiple candidate disaster types are then subjected to safety assessments in sequence according to the warning labels. Specifically, establish the correlation between coal mine disaster types and joint common parameters: Four types of coal mine disasters are configured: gas accident (X), roof fall accident (Y), water disaster (Z), and coal dust accident (W); six common parameters are set: A (gas concentration), B (roof pressure), C (temperature), D (humidity), E (wind speed), and F (coal dust concentration). Establish a mapping relationship between subtype libraries and parameters: Subtype library X (gas accident): parameters A, B, C; Subtype library Y (roof accident): parameters B, C, D; Subtype library Z (flood): parameters D, E, F; Subtype library W (coal dust accident): parameters A, E, F; Configure corresponding nodes for each parameter to form a node network: node A1 (parameter A), node B1 (parameter B), node C1 (parameter C), node D1 (parameter D), node E1 (parameter E), node F1 (parameter F); Collect data of each parameter in real time through a sensor network, and identify abnormal parameters based on preset conditions. Assume that abnormalities are detected in parameters C (temperature), D (humidity), and E (wind speed), and the corresponding abnormal nodes are C1, D1, and E1; Activate the sensing points of the corresponding abnormal nodes in each subtype library through the sensing chain, and perform differentiated marking: For abnormal node C1: activate the sensing point in subtype library X, mark information point C with a red circle, activate the sensing point in subtype library Y, and mark information point C with a red circle. For abnormal node D1: activate the sensing point in subtype library Y, mark information point D with a yellow triangle, activate the sensing point in subtype library Z, and mark information point D with a yellow triangle. For abnormal node E1: activate the sensing point in subtype library Z, mark information point E with a blue square, activate the sensing point in subtype library W, and mark information point E with a blue square. The marker information points in each subtype library are statistically analyzed: Subtype library X (gas accident): 1 marker information point (red circle-C); Subtype library Y (roof accident): 2 marker information points (red circle-C, yellow triangle-D); Subtype library Z (flood): 2 marker information points (yellow triangle-D, blue square-E); Subtype library W (coal dust accident): 1 marker information point (blue square-E); Therefore, the identified candidate disaster types include: gas accident (X), roof accident (Y), flood (Z), and coal dust accident (W); The accidents are sorted according to the number of marked information points: roof collapse accident (Y): 2 marked information points, flood disaster (Z): 2 marked information points, gas accident (X): 1 marked information point, coal dust accident (W): 1 marked information point; for roof collapse accident (Y) and flood disaster (Z) with the same number of marked information points, a secondary sort is made according to the importance of the marking method. Assuming the importance weights of the marking methods are: red circle (weight 1.0) > yellow triangle (weight 0.8) > blue square (weight 0.6), the weighted scores are calculated as follows: Roof collapse accident (Y): red circle × 1 + yellow triangle × 0.8 = 1.8 points; Flood (Z): yellow triangle × 0.8 + blue square × 0.6 = 1.4 points; The final priority ranking is: Roof collapse accident (Y) -1.8 points; Flood (Z) -1.4 points; Gas accident (X) -1.0 points; Coal dust accident (W) -0.6 points; Based on the ranking results, subsequent safety risk probability assessments are conducted using early warning labels and early warning models. The effective utilization of multi-parameter correlations cannot accurately identify potential disaster types in the early stages of a disaster.

[0030] S3: Input the warning label into the pre-built warning model to obtain the identification results containing the disaster risk probability distribution corresponding to each candidate disaster type. When the risk probability of any candidate disaster type exceeds the corresponding preset threshold, the warning signal is automatically triggered. The steps of inputting warning labels into a pre-built warning model to obtain the disaster risk probability distribution corresponding to each candidate disaster type, and automatically triggering a warning signal when the disaster risk probability of any candidate disaster exceeds the corresponding preset threshold include: inputting warning labels into a pre-built warning model; calculating the identification result containing the disaster risk probability distribution corresponding to each candidate disaster type through the warning model; monitoring the disaster risk probability corresponding to each candidate disaster type; and automatically triggering the corresponding warning signal when the disaster risk probability of any candidate disaster type exceeds its corresponding preset threshold.

[0031] It should be noted that the preset threshold is set independently for each disaster type and is dynamically adjusted based on at least one of the following factors: historical early warning accuracy, current production environment risk level, seasonal influencing factors, and equipment operating status. The automatic triggering of early warning signals includes the following steps: identifying all disaster risk probabilities exceeding preset thresholds; determining the corresponding disaster type and its probability value; determining the early warning level based on the probability value; generating an early warning signal containing the disaster type, probability value, and early warning level; and pushing the early warning signal to designated terminal devices. The early warning level is divided based on probability value intervals, including the following levels: low risk: probability value in the first interval; medium risk: probability value in the second interval; high risk: probability value in the third interval; and extremely high risk: probability value in the fourth interval. Through automated probability calculation and threshold comparison, intelligent and real-time early warning of coal mine disasters is achieved, effectively improving the response speed and accuracy of coal mine safety management. The steps for constructing the early warning model are as follows: establishing a mapping relationship between early warning labels and disaster types; calculating the prior probability P(disaster type) and likelihood P(early warning label|disaster type) based on historical data; calculating the posterior probability P(disaster type|early warning label) according to Bayes' theorem; generating a disaster risk probability distribution based on the posterior probability; the prior probability P(disaster type) is determined based on at least one of the following factors: historical disaster statistics, seasonal influencing factors, geological conditions, and mining technology characteristics; the likelihood P(early warning label|disaster type) is obtained by: analyzing the characteristics of monitoring data before historical disasters; statistically analyzing the frequency of early warning labels corresponding to each disaster type; establishing a conditional probability relationship database between early warning labels and disaster types; the posterior probability is calculated using the following Bayes' theorem: ,in, Indicates the first Types of disasters Indicates a warning label. Represents the posterior probability. Indicates likelihood. The prior probability is represented by the disaster risk probability distribution, which is expressed as a probability vector: {Disaster type 1: P1, Disaster type 2: P2, ..., Disaster type n: Pn}, where, ; Specifically, model parameters are established: prior probabilities P(hazard type), P(gas accident) = 0.40; P(flood) = 0.35; P(roof accident) = 0.25; for likelihoods P(warning label | hazard type), P(temperature increase | gas accident) = 0.80; P(methane concentration increase | gas accident) = 0.90; P(humidity increase | flood) = 0.85; P(roof pressure increase | roof accident) = 0.75; when a warning label B = {temperature increase, methane concentration increase} is received: calculate P(gas accident | B): ,in, , ,therefore, It outputs a judgment result containing the risk probabilities of all candidate disaster types. For example, the disaster risk probability distribution is: {"Gas accident": 0.878, "Flood": 0.107, "Roof collapse accident": 0.015}. Instead of waiting for a single indicator to exceed a threshold, it automatically triggers a warning signal of the corresponding level when the posterior probability of any disaster exceeds a set probability threshold, thus achieving a leap from "threshold alarm" to "probability prediction." By using Bayes' theorem to comprehensively evaluate multiple weak precursor signals when the data does not reach a fixed threshold, it calculates the potential risk probability of various disasters, thereby achieving earlier and more intelligent early warning.

[0032] S4: Based on the coal mine disaster type indicated by the early warning signal, which includes at least the disaster type, disaster location, and disaster level, generate a corresponding three-dimensional disaster evolution scenario. The scenario includes at least one of flood, gas outburst, and rock burst. In the generated disaster evolution scenario, configure emergency rescue drill tasks or safety training modules. The tasks include at least one of command drills, rescue drills, and self-rescue drills. Receive and respond to external user instructions, drive the three-dimensional scenario to change in response to user operations, execute interactive drills or training processes according to user instructions, and output drill results or training feedback. Specifically, the 3D virtual simulation engine is built on Unity3D or similar 3D modeling tools, supporting real-time lighting, material rendering, dynamic model loading, and scene switching. The disaster evolution scenario includes at least one of the following scene elements: the environmental state before the disaster, the dynamic evolution during the disaster, the simulation of the consequences after the disaster, and a visual demonstration of disaster prevention measures. The task configuration module supports automatically matching preset drill scripts according to the disaster type. These scripts include step sequences, role assignments, equipment operation procedures, and safety regulations. The user interaction module supports at least one of the following interaction methods: immersive operation with VR devices, mouse / keyboard / touchscreen interaction, voice command recognition and control, multi-user collaborative operation, and role-playing. It executes interactive drills or training processes according to user instructions and outputs results. The exercise results or training feedback also include: recording user operation behaviors and time points; scoring user operations according to preset scoring rules; outputting exercise reports, error messages, and improvement suggestions; in addition, a data interface module is set up for data docking with the coal mine intelligent disaster joint monitoring system to obtain monitoring data in real time; a model update module is used to update the disaster evolution model and exercise content according to actual disaster cases or new early warning rules; the generation of disaster evolution scenarios includes: calling a pre-built 3D model library; loading the corresponding physics engine parameters according to the disaster type; dynamically adjusting the disaster evolution path based on real-time monitoring data; exercise tasks include: users as commanders formulating rescue plans, users as rescuers performing equipment operations and personnel search and rescue, and users as underground personnel performing disaster avoidance and escape operations; Supported by a 3D environment model and scene editing system, a mine simulation system was created. The simulation system includes a realistic mine walkthrough, disaster prevention visualization, disaster emergency rescue and response, and mine substation power outage drills. Under normal circumstances, it serves as a monitoring, early warning, and interactive learning and training platform; under disaster circumstances, it serves as an emergency rescue auxiliary decision-making and command platform.

[0033] A coal mine safety assessment system based on large model empowerment, applied to any one of the above-mentioned coal mine safety assessment methods based on large model empowerment, comprising: The relationship building module is used to determine multiple coal mine disaster types and multiple joint common parameters, and to establish the correlation between multiple joint common parameters and multiple coal mine disaster types; The determination module is used to determine the joint common parameters that are abnormal based on preset conditions as an abnormal common parameter group, generate early warning labels based on the abnormal common parameter group, and determine the coal mine disaster type corresponding to the abnormal common parameter group based on the correlation to obtain at least one candidate disaster type; The risk assessment module is used to input the warning labels into the pre-built warning model to obtain the identification results containing the disaster risk probability distribution corresponding to each candidate disaster type. When the risk probability of any candidate disaster type exceeds the corresponding preset threshold, the warning signal is automatically triggered. The scenario simulation module is used to generate a corresponding three-dimensional scenario of disaster evolution based on the coal mine disaster type indicated by the early warning signal. In the generated disaster evolution scenario, an emergency rescue drill or safety training module is configured. By receiving and responding to external user instructions, the module drives the three-dimensional scenario to change in response to user operations, executes interactive drills or training processes according to user instructions, and outputs drill results or training feedback.

[0034] Intelligent disaster safety assessment and prevention in coal mines integrates data from safety monitoring, personnel location tracking, hydrological monitoring, fire monitoring, coal dust monitoring, and mine pressure monitoring to form a joint monitoring system. Real-time data from each subsystem can be viewed on a large screen. By establishing correlations between disasters and common joint data, comprehensive horizontal monitoring and early warning of various disasters such as water, fire, gas, and mine pressure are achieved, analyzing future trends and improving the accuracy of disaster early warning. A unified data joint monitoring platform is built to collect real-time sensing data from mine safety monitoring, personnel location tracking, water hazard monitoring, and rockburst monitoring, enabling unified data connection, management, and centralized display. It also performs statistical analysis on historical data to improve the coal mine safety supervision agencies' ability to monitor and perceive disaster risks such as water, fire, and gas in coal mines. It achieves business functions such as standardized interfaces, centralized monitoring, and timely alarms. The coal mine intelligent disaster prevention and control joint monitoring system connects to various business data through standardized interfaces. Through data collection, storage, and analysis, it ultimately displays real-time and historical data changes of joint monitoring on the monitoring platform page. Business personnel only need to monitor safety, personnel, fire, hydrology, coal dust, and mine pressure information through the homepage. It comprehensively presents various information underground in a three-dimensional scene, realizing intelligent mine geographic information, mine structure, and life... The system provides a real-time monitoring and simulation of various underground disaster monitoring systems, offering real-time warnings and simulations of major faults and hazards in each system. It also enables real-time simulation and linkage between different systems. Real-time monitoring of various toxic and harmful gases and working conditions at the working face in coal mines, such as methane content, carbon monoxide content, oxygen concentration, wind speed, negative pressure, and temperature (depending on the model of the gas monitoring system equipped in each mine), is provided. Automatic alarms and prompts are issued for any abnormal conditions. Monitoring data is displayed in real-time based on the raw data; when monitoring data exceeds the normal range, the page displays data with a red background, along with the alarm data range and the current actual data. Audio prompts indicate when one or more monitoring data points at the current working face have triggered an alarm. Real-time monitoring of personnel information, personnel exceeding time limits, overcrowding in key areas, attendance records, and personnel movement trajectories is also provided. It can promptly and accurately reflect the distribution information and movement trajectory of personnel underground to the dispatch center, so as to enable more reasonable scheduling and management; it can also conduct real-time and dynamic monitoring and early warning of water hazards that may be encountered during mining operations, such as aquifers and water-filled areas in old workings, in order to prevent water-related accidents.The mine hydrological monitoring data is networked to monitor important parameters such as water pressure, flow rate, water temperature, and water chemical parameters from hydrological observation wells or sensors in real time, enabling automatic alarm functions for exceeding limits. Scene-based monitoring is implemented according to the actual working face conditions, marking the locations of fire monitoring sensors on the working face. Monitoring information includes nitrogen, carbon dioxide, carbon monoxide, oxygen, and methane concentrations. When the monitored data exceeds the normal range, the corresponding data is displayed in red, along with the safe range and current value. Scene-based monitoring is also implemented according to the actual working face conditions, displaying coal dust monitoring markers and data from dust monitoring equipment at the actual monitoring locations. When the coal dust concentration exceeds the reasonable range, the data is displayed in red, along with the dust range and current dust concentration data. Changes in support height and support pressure are monitored. When the monitored data exceeds the normal range, the corresponding data is displayed in red, along with the safe range and current value. Once the data returns to normal, the display returns to normal. Potential joint common monitoring items are identified from a large number of coal mine disaster data sources. Joint common monitoring items are mainly extracted from various disasters. For example, predictions of flooding accidents include: dampness and darkening of the coal seam; increase in harmful gases; decreased temperature and cool coal walls; "sweat" on the coal walls; roof pressure and floor bulging; "red" iron oxide on the coal walls; water-like sounds; and the appearance of fog. From these, joint common monitoring items (temperature, humidity, gas, carbon monoxide, and roof pressure) can be extracted. Predictions of gas accidents sometimes occur in July and August, with methane concentrations of 5%–6% and oxygen concentrations not lower than 12%; sudden increases in heat source temperature, etc. From these, joint common monitoring items (date, temperature, and methane concentration) can be extracted. Predictions of coal dust accidents generally occur in winter; coal... Dust accumulation, excessive floating dust, and rising frictional temperature during equipment operation are all factors that can be analyzed to identify common parameters (date, temperature). Early warning models can then be used to predict the types of coal mine disasters corresponding to these abnormal common parameters, making a probabilistic prediction before a disaster is imminent. For example, if a set of mine pressure monitoring values ​​is collected from a data warehouse, the probability of a push-down roof collapse is 46%, a crushing roof collapse is 32%, and a flood is 22%. If the mine pressure value does not trigger a threshold, the prediction relies primarily on the experience of the monitoring personnel. If we express this using the conditional probability P(B|Ai), where P(temperature decrease | flood) = 22%, this method of calculating the probability of a category characteristic by predicting the category is called likelihood. Based on certain characteristics of the event, a probabilistic prediction is made. Each coal mine disaster has its own warning signs; for example, floods, push-down roof collapses, and crushing roof collapses all have different characteristics. When the joint monitoring items do not reach a threshold, a system threshold cannot be used for judgment. However, when the joint monitoring items reach a certain range, a probabilistic prediction can be made based on their respective predictive characteristics. This established the correlation between the target problem and data sources of different types and scales, and analyzed the degree of correlation between the data.

[0035] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0036] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A coal mine safety assessment method based on a large model, characterized in that, Includes the following steps: Identify multiple coal mine disaster types and multiple joint common parameters, and establish the correlation between multiple joint common parameters and multiple coal mine disaster types; Based on preset conditions, joint common parameters that are abnormal are identified as abnormal common parameter groups. Early warning labels are generated based on abnormal common parameter groups. Based on the correlation, the coal mine disaster type corresponding to the abnormal common parameter group is determined to obtain at least one candidate disaster type. The warning label is input into the pre-built warning model to obtain the identification result containing the disaster risk probability distribution corresponding to each candidate disaster type. When the risk probability of any candidate disaster type exceeds the corresponding preset threshold, the warning signal is automatically triggered. Based on the coal mine disaster type indicated by the early warning signal, a corresponding three-dimensional disaster evolution scenario is generated. In the generated disaster evolution scenario, an emergency rescue drill or safety training module is configured. By receiving and responding to external user instructions, the module drives the three-dimensional scenario to change in response to user operations, executes interactive drills or training processes according to user instructions, and outputs drill results or training feedback.

2. The coal mine safety assessment method based on a large model as described in claim 1, characterized in that: The steps for determining multiple coal mine disaster types and multiple joint common parameters include: Identify multiple types of coal mine disasters and configure a joint common parameter cluster for each type of coal mine disaster; Multiple joint common parameter clusters were obtained based on various coal mine disaster types; Obtain all joint common parameters from multiple joint common parameter groups, and extract all non-repeating joint common parameters to obtain multiple joint common parameters.

3. The coal mine safety assessment method based on a large model as described in claim 1, characterized in that: The steps for establishing the correlation between multiple joint common parameters and various coal mine disaster types include: Multiple coal mine disaster types and several common parameters are obtained and stored in a type database and a parameter database, respectively. The type database is divided into multiple subtype databases, each of which corresponds to a type of coal mine disaster. Multiple nodes are set up in the parameter database and connected to form a node network, where each node corresponds to a joint common parameter; Configure a data chain for the node, the data chain containing multiple data points arranged in a time series; Establish a sensing chain between the subtype library and the node network to form a correlation between coal mine disaster types and joint common parameters.

4. The coal mine safety assessment method based on a large model as described in claim 3, characterized in that: The steps for establishing the sensor link between the subtype library and the node network include: Multiple information points are configured for each subtype library, and the parameters in the joint common parameter group corresponding to each subtype library are distributed and stored in the information points. Configure a corresponding sensing point for each information point, and establish a sensing chain between sensing points and nodes that store the same joint common parameters.

5. The coal mine safety assessment method based on a large model as described in claim 1, characterized in that: The step of determining the joint common parameters that are abnormal based on preset conditions as an abnormal common parameter group, and generating warning labels based on the abnormal common parameter group includes: Configure preset conditions for each joint common parameter, collect real-time monitoring data of the joint common parameter in the coal mine, and take the joint common parameter corresponding to the real-time monitoring data that does not meet the corresponding preset conditions as the abnormal common parameter. Based on the synchronicity of parametric anomaly time series, multiple common anomaly parameters are combined to obtain anomaly common parameter clusters; Extract the feature information of the abnormal common parameter clusters and generate corresponding early warning labels.

6. The coal mine safety assessment method based on a large model as described in claim 1, characterized in that: The step of determining the coal mine disaster type corresponding to the abnormal common parameter cluster based on the correlation relationship to obtain at least one candidate disaster type includes: Obtain multiple common abnormal parameters and corresponding abnormal nodes from at least one abnormal common parameter group, and activate the sensing points of the corresponding abnormal nodes in each subtype library based on the sensing chain. In the subtype library, the information points corresponding to the sensing points are differentiated to obtain the marked information points. Among them, the information points of the same abnormal node in different subtype libraries are marked in the same way, and the information points corresponding to different abnormal nodes are marked in different ways. The disaster type corresponding to the subtype library with marked information points is used as the candidate disaster type.

7. The coal mine safety assessment method based on a large model as described in claim 1, characterized in that: The step of inputting early warning labels into a pre-built early warning model to obtain the disaster risk probability distribution corresponding to each candidate disaster type, and automatically triggering an early warning signal when the probability of any disaster risk exceeds the corresponding preset threshold, includes: Input the warning labels into the pre-built warning model; The early warning model is used to calculate the identification results, which include the probability distribution of disaster risk corresponding to each candidate disaster type. Monitor the disaster risk probability corresponding to each candidate disaster type, and automatically trigger the corresponding early warning signal when the disaster risk probability of any candidate disaster type exceeds its corresponding preset threshold.

8. A coal mine safety assessment system based on a large model, applied to the coal mine safety assessment method based on a large model as described in any one of claims 1-7, characterized in that, include: The relationship building module is used to determine multiple coal mine disaster types and multiple joint common parameters, and to establish the correlation between multiple joint common parameters and multiple coal mine disaster types; The determination module is used to determine the joint common parameters that are abnormal based on preset conditions as an abnormal common parameter group, generate early warning labels based on the abnormal common parameter group, and determine the coal mine disaster type corresponding to the abnormal common parameter group based on the correlation to obtain at least one candidate disaster type; The risk assessment module is used to input the warning labels into the pre-built warning model to obtain the identification results containing the disaster risk probability distribution corresponding to each candidate disaster type. When the risk probability of any candidate disaster type exceeds the corresponding preset threshold, the warning signal is automatically triggered. The scenario simulation module is used to generate a corresponding three-dimensional scenario of disaster evolution based on the coal mine disaster type indicated by the early warning signal. In the generated disaster evolution scenario, an emergency rescue drill or safety training module is configured. By receiving and responding to external user instructions, the module drives the three-dimensional scenario to change in response to user operations, executes interactive drills or training processes according to user instructions, and outputs drill results or training feedback.

Images