A mine digital gas parameter early warning system based on gas analysis

Abstract

The present application relates to the technical field of mine safety monitoring, and discloses a mine digital gas parameter early warning system based on gas analysis. The system comprises data acquisition and feature extraction, spatial topology correlation, gas state intention analysis, risk deduction based on a knowledge graph, and hierarchical early warning generation, which realizes intelligent monitoring of mine gas safety. The gas behavior mode is dynamically correlated with the roadway space structure, forming networked situation awareness, and the multi-source precursor information is deeply semantically correlated and probabilistically inferred using a disaster knowledge graph, thereby generating a comprehensive risk conclusion with causal logic. The system realizes the leap from discrete point alarm to spatial situation analysis, from simple threshold judgment to intelligent causal deduction, and improves the accuracy of early warning and decision support capability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mine safety monitoring, in particular to a mine digital gas parameter early warning system based on gas analysis. BACKGROUND

[0002] Mine disaster early warning mainly relies on the gas sensor network deployed in the roadway. The existing technical solution is usually to independently monitor the original gas concentration data collected by each sensor, and when any data exceeds the preset static threshold, local alarm is triggered. This single-point threshold-based mode regards each sensor as an information island, and its alarm only reflects the instantaneous state of the space point, lacking analysis of the spatial correlation and dynamic propagation law of gas anomalies in the roadway network. The precursor of a gas disaster is often a complex process of multi-factor coupling and gradual evolution in space, and the existing point monitoring is difficult to form a coherent understanding of the overall roadway gas situation.

[0003] The risk judgment of the prior art mainly relies on a simple "if-then" rule base or a model based on historical statistics. These methods deal with explicit and single-dimensional signal correlation, and are not suitable for complex scenarios that require deep reasoning of roadway topology, environmental parameters, multi-gas interaction and disaster evolution knowledge. The rule base is difficult to exhaust all hidden coupling situations, and cannot dynamically simulate the propagation path of the risk, resulting in limited accuracy of the early warning, high false alarm rate, and the early warning information provided is mainly phenomenon description, lacking the conclusion of disaster causes, development path and influence range. An early warning method is needed that can understand the spatial semantics of gas state and perform intelligent correlation reasoning. SUMMARY

[0004] The purpose of the present application is to provide a mine digital gas parameter early warning system based on gas analysis to solve the problems raised in the background art.

[0005] To achieve the above purpose, the present application provides a mine digital gas parameter early warning system based on gas analysis, which comprises:

[0006] A data acquisition and feature extraction module acquires the original gas data stream and auxiliary environmental data stream generated by various sensors distributed in the mine roadway, extracts the features of the original gas data stream and auxiliary environmental data stream, and forms a gas state behavior pattern;

[0007] A spatial correlation module correlates the gas state behavior pattern with the roadway spatial topology structure according to the mine geographic information model, and generates a gas state behavior pattern carrying a spatial label;

[0008] an intent analysis module configured to perform intent matching and decomposition on the gas state behavior pattern with the space label based on a predefined gas state intent type library, to generate a plurality of basic gas state intent segments for a specific roadway, and to assign a confidence score to each basic gas state intent segment to build an intent segment set;

[0009] a risk deduction module configured to call a predefined mine disaster knowledge graph, to find a matching associated rule for each basic gas state intent segment in the intent segment set, to activate associated graph nodes and edges to form an initial risk deduction path, and to perform spatiotemporal consistency verification and probability integration on the initial risk deduction path to generate one or more comprehensive risk deduction conclusions;

[0010] an early warning generation module configured to generate a graded early warning instruction based on the emergency degree and influence range of the comprehensive risk deduction conclusion.

[0011] Preferably, the feature extraction on the original gas data stream and the auxiliary environmental data stream to form the gas state behavior pattern comprises:

[0012] synchronization and time sequence alignment of the original gas data stream and the auxiliary environmental data stream to generate a standardized fusion data sequence;

[0013] extraction of a gas concentration change rate, an environmental parameter covariant trend, and a historical abnormal segment recurrence feature from the fusion data sequence to form the gas state behavior pattern; the specific steps include:

[0014] calculation of a change gradient of the gas concentration data in the fusion data sequence in a unit time by using a sliding time window, and recording the change gradient as the gas concentration change rate;

[0015] synchronous analysis of temperature and humidity data in the fusion data sequence, calculation of a covariance matrix of the temperature and humidity data and the gas concentration change rate, and determination of the environmental parameter covariant trend according to a principal component direction of the covariance matrix;

[0016] retrieval of a historical gas database, calling of historical data with a similar space label as the fusion data sequence, comparison of the current sequence and historical abnormal segments by using a dynamic time warping algorithm, calculation of a similarity degree, and marking of a similarity degree result exceeding a threshold value as a historical abnormal segment recurrence feature;

[0017] normalization splicing of the gas concentration change rate, the environmental parameter covariant trend, and the historical abnormal segment recurrence feature to form a multi-dimensional feature vector, and the multi-dimensional feature vector is the gas state behavior pattern.

[0018] Preferably, the step of associating the gas state behavior pattern with the roadway space topology according to the mine geographic information model comprises:

[0019] Reading a mine geographic information model, which defines roadway nodes and connecting edges in a graph structure;

[0020] Mapping the physical location of the sensor corresponding to the gas state behavior pattern to the nearest roadway node in the mine geographic information model;

[0021] According to the mapping relationship, adding the unique identifier of the roadway node and the identifier list of adjacent nodes to the gas state behavior pattern to generate a gas state behavior pattern carrying a space tag;

[0022] The gas state behavior pattern carrying a space tag not only contains gas characteristics, but also implies its propagation and influence potential in the roadway network.

[0023] Preferably, the step of performing intent matching and decomposition on the gas state behavior pattern carrying a space tag based on a pre-defined gas state intent type library comprises:

[0024] The gas state intent type library enumerates a plurality of independent gas abnormal states, and each gas abnormal state is defined by a set of feature condition rules;

[0025] The multi-dimensional feature vector in the gas state behavior pattern carrying a space tag is compared with the feature condition rules of each gas abnormal state in the gas state intent type library one by one;

[0026] When the multi-dimensional feature vector satisfies part or all of the feature condition rules of a certain gas abnormal state, it is determined that the gas abnormal state is contained in the gas state behavior pattern carrying a space tag, and a basic gas state intent segment is generated;

[0027] One gas state behavior pattern carrying a space tag matches and decomposes multiple basic gas state intent segments, and each basic gas state intent segment represents a potential abnormal type identified.

[0028] Preferably, the step of assigning a confidence score to each basic gas state intent segment to construct an intent segment set comprises:

[0029] For each generated basic gas state intent segment, retrieve the feature condition rules on which it is matched;

[0030] Count the number of items in the multi-dimensional feature vector that satisfy the feature condition rules, and calculate the proportion of the number of satisfied items in the total number of rule items as the rule satisfaction degree;

[0031] The rule satisfaction degree is weighted and corrected according to the similarity value of the historical abnormal segment corresponding to the basic gas state intention picture segment, to obtain an initial confidence degree;

[0032] The initial confidence degree is secondarily adjusted according to the activity degree of the roadway node where the basic gas state intention picture segment is located in historical disasters, to generate a final confidence score;

[0033] All the basic gas state intention picture segments and the corresponding confidence scores jointly constitute an intention picture segment set.

[0034] Preferably, the step of calling a predefined mine disaster knowledge graph to find a matching associated rule for each basic gas state intention picture segment in the intention picture segment set comprises:

[0035] The mine disaster knowledge graph is parsed, wherein the nodes in the mine disaster knowledge graph represent geological entities, gas states or disaster events, and the edges represent causal relationships or accompanying relationships;

[0036] The basic gas state intention picture segment is taken as a query input to find a node with the basic gas state intention picture segment name as the core in the mine disaster knowledge graph;

[0037] Starting from the found core node, all directly connected edges and adjacent nodes are traversed, and these edges and adjacent nodes constitute an associated rule directly related to the basic gas state intention picture segment;

[0038] The activated nodes and edges in the traversal are recorded as a local reasoning subgraph of the basic gas state intention picture segment.

[0039] Preferably, the step of activating associated graph nodes and edges to form an initial risk deduction path comprises:

[0040] For all basic gas state intention picture segments in the intention picture segment set, the local reasoning subgraphs activated by the basic gas state intention picture segments are superimposed;

[0041] In the superimposed graph, nodes and edges activated by multiple local reasoning subgraphs are identified, and the nodes and edges become high-risk convergence points;

[0042] Taking the high-risk convergence point as the core, the high-risk convergence point is traced back and connected to each basic gas state intention picture segment node as a query starting point, to form one or more reasoning chains from the intention picture segment to potential disaster events, and the reasoning chains are the initial risk deduction path;

[0043] Each of the initial risk deduction paths is accompanied by a path strength value, which is calculated by the confidence scores of all the basic gas state image segments constituting the path and the weights of the edges.

[0044] Preferably, the step of performing spatio-temporal consistency check and probability integration on the initial risk deduction paths comprises:

[0045] Obtaining the spatial position information of all the roadway nodes involved in the initial risk deduction paths;

[0046] Checking the connectivity of adjacent nodes in the path in the mine geographic information model, and if the logically deduced path does not match the actual roadway connectivity, the path strength value of the initial risk deduction path is reduced;

[0047] Introducing a time decay factor, and for the initial risk deduction paths generated based on earlier time data, the path strength value thereof decays over time;

[0048] Classifying all the initial risk deduction paths that pass the spatio-temporal check according to the types of the final disaster events to which the paths point;

[0049] Performing probability fusion calculation on all the path strength values pointing to the same disaster event type to generate a comprehensive risk probability value, and the comprehensive risk probability value and the description of the disaster event corresponding thereto jointly constitute a comprehensive risk deduction conclusion;

[0050] The step of formatting the graded early warning instruction comprises:

[0051] Predefining a risk probability threshold and an emergency level table for different types of disaster events;

[0052] Comparing the comprehensive risk probability value in the comprehensive risk deduction conclusion with the risk probability threshold to determine the emergency level thereof;

[0053] Determining the influence range level of the comprehensive risk deduction conclusion according to the range of the roadway nodes affected thereby;

[0054] Selecting a corresponding instruction template from an early warning instruction template library in combination with the emergency level and the influence range level;

[0055] Filling the specific description of the comprehensive risk deduction conclusion and the involved roadway position information into the selected instruction template to generate a final graded early warning instruction.

[0056] Preferably, the step of constructing the mine geographic information model comprises:

[0057] The point cloud data of a mine roadway is acquired by a three-dimensional laser scanner, and design drawings and exploration data of the roadway are synchronously imported, the point cloud data is denoised and registered, and an accurate three-dimensional spatial point cloud model of the roadway is generated;

[0058] Based on the three-dimensional spatial point cloud model of the roadway, a spatial clustering and feature recognition algorithm is used to segment and extract point cloud sets of each independent roadway, and then a center axis and a cross section profile of each roadway are fitted;

[0059] The intersection of the center axis of each roadway at both ends and other axes is defined as a node, and the center axis segment connecting two nodes is defined as an edge, so that the roadway network is abstracted into a graph structure composed of nodes and edges;

[0060] The three-dimensional coordinates of each node in the graph are calculated, and the length, slope, azimuth and connected upstream and downstream node identifiers of each edge are calculated, and the three-dimensional coordinates, length, slope, azimuth and node identifiers are attached to the corresponding graph elements as attributes;

[0061] The roadway topology graph with attributes is associated and encapsulated with the three-dimensional spatial point cloud model of the roadway, to form a mine geographic information model containing both geometric models and topological relationships.

[0062] Preferably, the construction steps of the mine disaster knowledge graph include:

[0063] From mine geological reports, safety procedure documents, historical accident case libraries and field expert experience manuals, the associated knowledge between gas anomalies, geological structures, mining activities and various mine disaster events is collected;

[0064] The ontology structure of the knowledge graph is defined, the core entity types and their attributes are determined, the core entity types include geological entities, gas states, environmental parameters, human activities and disaster events, the relationship types between entities are defined, and the relationship types include causal relationship, accompanying relationship, triggering relationship and inhibition relationship;

[0065] For structured historical case data, entity attributes and relationships are directly mapped, and "reason-event-result" triples are extracted;

[0066] For unstructured text reports and procedures, named entity recognition and relationship extraction are performed by using natural language processing technology, and a graph substructure with entities as nodes and relationships as edges is extracted;

[0067] The entities, relationships and attributes extracted from all sources are aligned, disambiguated and fused under a unified ontology framework to construct a mine disaster knowledge graph.

[0068] Compared with the prior art, the present application has the advantages that:

[0069] By spatially labeling the extracted gas state behavior patterns based on the roadway spatial topology in the mine's geographic information model, each identified gas behavior pattern carries its specific location and connectivity information within the roadway network. This allows the system to understand that gas anomalies or trends are not isolated events, but rather dynamic processes closely tied to the physical spatial structure with potential propagation paths. Traditional discrete-point alarms are integrated into a dynamic situation map covering the roadway network and labeled with gas behavior characteristics, achieving a leap from "point-based perception" to "networked spatial situational awareness," providing accurate spatial context for analyzing gas diffusion, accumulation, and migration.

[0070] By invoking a predefined graph containing knowledge of mine disaster mechanisms and cases, the system identifies and activates associated disaster nodes, trigger nodes, and logical relationship edges within the graph for each identified and spatially labeled gas behavior intent fragment, thereby dynamically generating multiple possible risk evolution paths. Subsequently, the spatiotemporal consistency of these paths is verified and probabilistically integrated to select the most reasonable comprehensive deduction conclusion. A reasoning engine simulating expert thinking is constructed, enabling the early warning system to possess deep semantic association and causal reasoning capabilities. Based on scattered, multi-point, and multi-type gas precursor information, it can automatically associate it with specific disaster types and deduce its possible causes, development stages, and future impact range. The output is a reasoned and structured risk report, rather than a simple exceedance alarm, improving the accuracy of the early warning and its decision support value. Attached Figure Description

[0071] Figure 1 This is a timing diagram of the mine digital gas parameter early warning system based on gas analysis described in this invention;

[0072] Figure 2 A flowchart for the spatial association module;

[0073] Figure 3 A flowchart for constructing a set of intent fragments;

[0074] Figure 4 A spatiotemporal verification line graph of risk probability for a mine gas parameter early warning system;

[0075] Figure 5 A pie chart showing the distribution of conditions in mine roadways. Detailed Implementation

[0076] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0077] Please refer to Figure 1 The present application provides a mine digital gas parameter early warning system based on gas analysis, which comprises a data acquisition and feature extraction module, a space correlation module, an intention analysis module, a risk deduction module and an early warning generation module. The overall implementation scheme of the system is as follows: various sensors deployed in the mine roadway continuously generate original gas data streams and auxiliary environmental data streams, the data acquisition and feature extraction module receives these data streams and performs feature extraction, thereby forming a gas state behavior pattern. The space correlation module correlates the gas state behavior pattern with the roadway space topology according to a pre-built mine geographic information model, thereby generating a gas state behavior pattern carrying a space label. The intention analysis module has a pre-defined gas state intention type library, which matches and decomposes the gas state behavior pattern carrying a space label, produces a number of basic gas state intention segments for a specific roadway, and assigns a confidence score to each segment to build an intention segment set. The risk deduction module calls a pre-defined mine disaster knowledge graph, finds a matching correlation rule for each basic gas state intention segment in the intention segment set, activates the associated graph nodes and edges to form an initial risk deduction path, and performs spatio-temporal consistency checking and probability integration on these initial risk deduction paths, finally generating one or more comprehensive risk deduction conclusions. The early warning generation module formats a graded early warning instruction according to the emergency degree and impact range of the comprehensive risk deduction conclusion, thereby completing a complete closed loop from data perception to risk early warning.

[0078] Embodiment 1: Please refer to Figure 2The data acquisition and feature extraction module synchronizes and time-aligns the original gas data stream and the auxiliary environmental data stream to generate a standardized fusion data sequence. The gas concentration rate of change, environmental parameter covariant trend, and historical abnormal segment recurrence feature are extracted from the fusion data sequence. The gas concentration rate of change is calculated using a sliding time window. The temperature and humidity data are simultaneously analyzed to calculate the covariance matrix of the gas concentration rate of change, and the principal component direction of the covariance matrix is determined. The historical abnormal segment recurrence feature requires searching the historical gas database, calling the historical data with similar spatial tags as the current fusion data sequence, comparing the current sequence with the historical abnormal segments using the dynamic time warping algorithm, and calculating the similarity. The similarity results exceeding the threshold are marked. The obtained gas concentration rate of change, environmental parameter covariant trend, and historical abnormal segment recurrence feature are normalized and spliced to form a multi-dimensional feature vector, which is the gas state behavior pattern. The spatial correlation module reads the mine geographic information model defined by the graph structure to define the roadway nodes and connection edges. The module maps the sensor physical location corresponding to the gas state behavior pattern to the nearest roadway node in the mine geographic information model. According to the mapping relationship, the unique identifier of the roadway node and the identifier list of adjacent nodes are attached to the gas state behavior pattern, thereby generating a gas state behavior pattern carrying a spatial tag. The gas state behavior pattern carrying a spatial tag not only contains gas features, but also implies the propagation and influence potential in the roadway network.

[0079] In a specific implementation, a sensor array deployed in the return airway of a mine coal mining face generates an original gas data stream and an auxiliary environmental data stream, the original gas data stream includes time series readings of methane concentration and carbon monoxide concentration, and the auxiliary environmental data stream includes time series readings of temperature and humidity. The data acquisition and feature extraction module synchronizes and time-aligns the two types of data streams to merge the four types of parameters with a unified timestamp to generate a standardized fusion data sequence. In a specific implementation, the module extracts the gas concentration rate of change, the environmental parameter covariant trend, and the historical abnormal segment recurrence feature from the standardized fusion data sequence. For the methane concentration data in the fusion data sequence, a sliding time window with a width of five minutes is used to calculate the linear regression slope of the concentration values within the window. The slope is recorded as the rate of change per unit time as the gas concentration rate of change. The rate of change of carbon monoxide concentration in the fusion data sequence is calculated in the same way. The temperature and humidity data in the fusion data sequence are simultaneously analyzed to calculate the covariance matrix between the temperature change sequence, the humidity change sequence, and the aforementioned methane concentration rate of change sequence. The environmental parameter covariant trend is determined according to the eigenvector corresponding to the first principal component direction of the covariance matrix. This trend represents the main coupling relationship between environmental parameters and gas concentration change.

[0080] In a specific implementation, a historical gas database associated with the current monitoring point is retrieved, historical data with the same "coal mining face return airway" spatial label as the current monitoring point is called, a dynamic time warping algorithm is used to compare the current five-minute fusion data sequence with multiple event fragments marked as "gas abnormal emission" in the historical database one by one, the similarity distance between the sequences is calculated, the distance value is converted into a similarity score between 0 and 1, and a similarity result exceeding the threshold of 0.8 is marked as a historical abnormal fragment recurrence feature. The gas concentration change rate, the principal component coefficient of the environmental parameter covariant trend, and the similarity score of the historical abnormal fragment recurrence feature are extracted, mapped to the [0, 1] interval through the maximum and minimum value normalization method, spliced in a fixed order after normalization, and a multi-dimensional feature vector is constructed. The multi-dimensional feature vector is the gas state behavior pattern at the current time for the monitoring point. In some embodiments, the gas concentration change rate calculation formula is expressed as

[0081] ;

[0082] wherein: denotes the concentration change rate of gas g, denotes the change amount of gas g concentration within the time window .

[0083] In some embodiments, the spatial correlation module reads a predefined mine geographic information model, the mine geographic information model is stored in a graph structure, a node represents a roadway intersection point or an endpoint, an edge represents a roadway segment connecting the nodes, and the module maps the physical location of the sensor generating the gas state behavior pattern to the nearest roadway node in the mine geographic information model according to the built-in coordinate correspondence table. It can be understood that if the sensor is located at the roadway position represented by the node numbered "N201", the mapping relationship is established. According to the mapping relationship, the spatial correlation module attaches a unique identifier of the roadway node "N201" to the gas state behavior pattern, and retrieves and attaches a list of identifiers of all adjacent nodes directly connected to "N201" from the mine geographic information model, such as "[N200, N202, N210]", thereby generating a gas state behavior pattern carrying a spatial label. Optionally, the gas state behavior pattern carrying a spatial label not only contains a multi-dimensional feature vector of gas features, but also implicitly contains the propagation and influence potential of the pattern in the roadway network through its attached spatial label. It can be understood that the influence of the pattern spreads along the topological path indicated by the adjacent node list in the label.

[0084] Embodiment 2: refer to Figure 3, the intent analysis module operates based on a predefined gas state intent type library, which enumerates multiple independent gas abnormal states, each defined by a set of feature condition rules. The intent matching and decomposition process compares the multi-dimensional feature vector in the spatially tagged gas state behavior pattern with the feature condition rules of each gas abnormal state in the gas state intent type library one by one. When the multi-dimensional feature vector satisfies part or all of the feature condition rules of a certain gas abnormal state, it is determined that this spatially tagged gas state behavior pattern contains this gas abnormal state, and a basic gas state intent segment is generated. A spatially tagged gas state behavior pattern can match and decompose multiple basic gas state intent segments at the same time, each segment representing a potential abnormal type identified. The process of assigning a confidence score to each basic gas state intent segment to build the intent segment set is as follows: for each generated basic gas state intent segment, retrieve the feature condition rules on which it is matched. Count the number of items in the multi-dimensional feature vector that satisfy these feature condition rules, and calculate the proportion of the number of satisfied items in the total number of rule items as the rule satisfaction degree. Combine the similarity value of the historical abnormal segment recurrence feature corresponding to the basic gas state intent segment to weight and correct the rule satisfaction degree to obtain the initial confidence. Adjust the initial confidence according to the activity level of the roadway node where the basic gas state intent segment is located in historical disasters to generate the final confidence score. All basic gas state intent segments and their corresponding confidence scores together constitute the intent segment set.

[0085] In a specific implementation, the intent analysis module operates based on a predefined gas state intent type library, which is stored in a structured list form and enumerates multiple independent gas abnormal states such as "slow gas accumulation", "early coal spontaneous combustion", and "gas overflow in goaf", each defined by a set of feature condition rules. For example, the feature condition rules of the "slow gas accumulation" state include "methane concentration change rate greater than zero for three consecutive periods", "absolute value of environmental temperature change rate less than threshold K1", and "no obvious carbon monoxide abnormal feature matched". In a specific implementation, a spatially tagged gas state behavior pattern from the roadway of the heading face is sent to the intent analysis module, and the multi-dimensional feature vector in the behavior pattern includes normalized methane concentration change rate 0.85, carbon monoxide concentration change rate 0.10, temperature covariant coefficient 0.30, and historical abnormal segment recurrence similarity 0.75. The intent matching and decomposition process compares the above multi-dimensional feature vector with the feature condition rules of each gas abnormal state in the gas state intent type library one by one.

[0086] In some embodiments, the comparison process finds that the multi-dimensional feature vector satisfies the condition of "methane concentration rate of change is greater than zero for three consecutive periods" in the feature condition rule of "slow gas accumulation" state, and also satisfies the condition of "absolute value of ambient temperature rate of change is less than threshold K1", but the condition of "no obvious carbon monoxide abnormal feature is matched" is also satisfied, so it is determined that the gas state behavior pattern carrying the space label contains the "slow gas accumulation" gas abnormal state, and a basic gas state intention segment named "slow gas accumulation_drilling face" is generated. It can be understood that the same multi-dimensional feature vector also partially satisfies the partial rule of "early coal spontaneous combustion", such as "detectable upward trend of carbon monoxide concentration", thereby simultaneously generating another basic gas state intention segment named "early coal spontaneous combustion_drilling face". One gas state behavior pattern carrying a space label matches and decomposes multiple basic gas state intention segments, and each basic gas state intention segment represents a potential abnormal type identified.

[0087] In a specific implementation, a confidence score is assigned to each basic gas state intention segment to construct an intention segment set. For the generated "slow gas accumulation_drilling face" basic gas state intention segment, the feature condition rule set on which it is matched is retrieved, which contains 5 sub-rules in total. The number of rule items in the multi-dimensional feature vector that are fully satisfied is 4, and the rule satisfaction degree is calculated to be 0.8. The similarity value of the historical abnormal segment reproduction feature corresponding to the basic gas state intention segment is 0.75. An initial confidence is obtained by using a predefined weight coefficient a to weight and correct the rule satisfaction degree. It can be understood that the weight coefficient a is used to adjust the influence degree of historical similarity on the current judgment, and the calculation formula is expressed as

[0088] ;

[0089] Wherein: represents the initial confidence of the basic gas state intention segment, represents the rule satisfaction degree, represents the similarity value of the historical abnormal segment reproduction feature, and a is a preset weight coefficient.

[0090] In some embodiments, the "slow gas accumulation_drilling face" basic gas state intention segment is adjusted again according to the activity degree of the roadway node in which it is located in the historical disaster case library. It is found that the node has been recorded as a "gas abnormality" related event 3 times in the past year. According to a preset activity degree-adjustment value mapping table, it is determined that the adjustment factor β is 0.1, and the initial confidence is adjusted to Adding this to the adjustment factor β generates a final confidence score of 0.92. Optionally, the "early stage of spontaneous combustion of coal in tunneling face" basic gas state intention segment is calculated with a confidence score of 0.65 based on its rule satisfaction, historical similarity, and the historical fire activity of its different nodes, using the same process. All basic gas state intention segments and their corresponding confidence scores together constitute an intention segment set, for example, the set content is {(slow gas accumulation in tunneling face, 0.92), (early stage of spontaneous combustion of coal in tunneling face, 0.65)}.

[0091] Example 3: The risk deduction module calls a predefined mine disaster knowledge graph. Nodes in this graph represent geological entities, gas states, or disaster events, while edges represent causal or accompanying relationships. For each basic gas state intent fragment in the intent fragment set, matching association rules are found. First, the basic gas state intent fragment is used as the query input, and the mine disaster knowledge graph searches for nodes with the name of that intent fragment as their core. Starting from the found core node, all directly connected edges and adjacent nodes are traversed. These edges and adjacent nodes constitute association rules directly related to that basic gas state intent fragment. The nodes and edges activated during the traversal are recorded as local reasoning subgraphs for that basic gas state intent fragment. The process of activating associated graph nodes and edges to form the initial risk deduction path is as follows: For all basic gas state intent fragments in the intent fragment set, their respective activated local reasoning subgraphs are superimposed. In the superimposed graph, nodes and edges jointly activated by multiple local reasoning subgraphs are identified; these nodes and edges become high-risk convergence points. Taking high-risk convergence points as the core, the system traces back and connects to various basic gas state intent fragment nodes that serve as the starting point for queries, forming one or more reasoning chains pointing from intent fragments to potential disaster events. This reasoning chain constitutes the initial risk projection path. Each initial risk projection path is accompanied by a path strength value, which is calculated jointly by the confidence scores of all basic gas state intent fragments constituting the path and the weights of the edges traversed along the path.

[0092] In a specific implementation, the risk deduction module calls a predefined mine disaster knowledge graph to process the set of intention segments from the intention analysis module, which contains three basic gas state intention segments, namely "abnormal gas concentration_ return airway", "carbon monoxide concentration slowly rising_ working face", and "temperature abnormally rising_ goaf boundary". The module analyzes the mine disaster knowledge graph, which is stored in a graph structure of nodes and edges. The nodes represent "goaf", "gas", "carbon monoxide", "coal seam", "gas explosion", "coal spontaneous combustion", gas state or disaster event, and the edges represent "cause", "accompany", and "easily occur". In a specific implementation, for each basic gas state intention segment in the set of intention segments, find the matching association rule. Taking the "abnormal gas concentration_ return airway" basic gas state intention segment as an example, it is used as a query input to search in the node index of the mine disaster knowledge graph, and the core node named "abnormal gas concentration" is found.

[0093] In a specific implementation, starting from the found "abnormal gas concentration" core node, all directly connected edges and adjacent nodes from and pointing to the node are traversed. During the traversal process, a "may cause" edge is connected to the "gas accumulation" node, a "often accompany" edge is connected to the "poor ventilation" node, and a "easily induce" edge is connected to the "gas explosion" node. These traversed edges "may cause", "often accompany", "easily induce" and adjacent nodes "gas accumulation", "poor ventilation", "gas explosion" together constitute the association rule directly related to the "abnormal gas concentration_ return airway" basic gas state intention segment. Record the activated "abnormal gas concentration", "gas accumulation", "poor ventilation", "gas explosion" four nodes and the three edges connecting these nodes. These nodes and edges constitute the local reasoning subgraph of the "abnormal gas concentration_ return airway" basic gas state intention segment. It can be understood that the same operation is performed on the "carbon monoxide concentration slowly rising_ working face" and "temperature abnormally rising_ goaf boundary" basic gas state intention segments in the set of intention segments, and their respective local reasoning subgraphs are activated.

[0094] In some embodiments, the associated graph nodes and edges are activated to form an initial risk deduction path, for all basic gas state scenario nodes in the scenario node set, the local reasoning subgraph activated by "gas concentration anomaly_ return airway", the local reasoning subgraph activated by "carbon monoxide concentration slow rise_ working face" and the local reasoning subgraph activated by "temperature anomaly rise_ goaf boundary" are superimposed in memory, the superimposed graph combines all activated nodes and edges, in the superimposed graph, nodes and edges activated by multiple local reasoning subgraphs are identified, it can be understood that the node "coal spontaneous combustion" appears in both the local reasoning subgraph of "carbon monoxide concentration slow rise_ working face" and the local reasoning subgraph of "temperature anomaly rise_ goaf boundary", and this node becomes a high-risk convergence point, the node "gas explosion" appears in the local reasoning subgraph of "gas concentration anomaly_ return airway", and also becomes a high-risk convergence point through the indirect association with the "high temperature" and "coal spontaneous combustion" subgraphs through the shared node.

[0095] In some embodiments, taking the "coal spontaneous combustion" high-risk convergence point as the core, the superimposed knowledge graph is traced back and connected to the "carbon monoxide concentration slow rise_ working face" basic gas state scenario node and the "temperature anomaly rise_ goaf boundary" basic gas state scenario node as the starting point of the query, forming a reasoning chain from the two scenario nodes to the "coal spontaneous combustion" disaster event, which is an initial risk deduction path, taking the "gas explosion" high-risk convergence point as the core, tracing back and connecting to the "gas concentration anomaly_ return airway" basic gas state scenario node, forming another initial risk deduction path, optionally, each initial risk deduction path is attached with a path strength value, the path strength value is calculated by the confidence scores of all basic gas state scenario nodes and the weights of the edges passing through, and the calculation formula is expressed as

[0096] ;

[0097] Wherein: represents the path strength value of the jth initial risk deduction path, represents the set of basic gas state scenario nodes constituting the path, represents the confidence score of the ith basic gas state scenario node, represents the set of edges passing through the path, represents the pre-defined weight of the kth edge in the knowledge graph, represents the size of the segment set for average calculation, it can be understood that the first path about "coal spontaneous combustion" involves two scenario nodes, and the path strength value calculation will include the confidence scores of 0.70 and 0.60 and the weights of the two edges passing through.

[0098] Referring toFigure 4 This is a spatiotemporal verification line chart of risk probability for a mining gas parameter early warning system. It corresponds to the spatiotemporal consistency verification stage in the "risk simulation phase," and is primarily used to assess the changes in risk probability and threshold compliance for different disaster paths. The risk probability initially rises (reaching a peak of 0.6 after approximately 2 hours), then continuously declines (falling below the threshold after 6 hours), only exceeding the threshold within a 2-4 hour range. The risk probability continuously rises, gradually approaching 0.8 from 0.2, remaining above the threshold throughout, showing a continuous accumulation trend. The risk probability of the gas explosion path better meets the conditions for "triggering an early warning," while the coal spontaneous combustion path only poses a risk within a short period. This type of chart is a key verification tool for risk simulation—by comparing risk probabilities with spatiotemporal thresholds, disaster paths requiring "priority warnings" can be identified, providing a core basis for generating tiered early warning instructions and ensuring the accuracy and timeliness of the warnings.

[0099] Example 4: The steps for spatiotemporal consistency verification and probability integration of the initial risk projection path include: obtaining the spatial location information of all roadway nodes involved in the initial risk projection path; checking the connectivity of adjacent nodes in the path in the mine geographic information model; if the logically inferred path does not match the actual roadway connectivity, reducing the path strength value of the initial risk projection path; introducing a time decay factor, where the path strength value of the initial risk projection path generated based on earlier time data decays over time; classifying all initial risk projection paths that pass spatiotemporal verification according to the final disaster event type they point to; performing probability fusion calculation on all path strength values ​​pointing to the same disaster event type to generate a comprehensive risk probability value, which, along with its corresponding disaster event description, constitutes a comprehensive risk projection conclusion. The steps for the early warning generation module to format and generate graded early warning instructions include: pre-setting risk probability thresholds and urgency level comparison tables for different categories of disaster events; comparing the comprehensive risk probability value in the comprehensive risk projection conclusion with the risk probability thresholds to determine its urgency level. Based on the scope of the affected roadway nodes according to the comprehensive risk simulation results, the impact level is determined. Combining the urgency level and the impact level, a corresponding instruction template is selected from the early warning instruction template library. The specific description of the comprehensive risk simulation results and the location information of the affected roadways are then filled into the selected instruction template to generate the final graded early warning instruction.

[0100] In a specific implementation, the risk deduction module performs spatio-temporal consistency verification and probability integration on the generated initial risk deduction path. Assuming there are two initial risk deduction paths, path A points to the "gas explosion" disaster event and involves roadway nodes "N101", "N102", and "N105", and path B points to the "coal spontaneous combustion" disaster event and involves roadway nodes "N201" and "N203". The spatial location information of all roadway nodes involved in path A and path B is obtained from the mine geographic information model. The logical connectivity of adjacent nodes in path A is checked. The reasoning chain of path A is "N101 -> N102 -> N105". It is found that there is no direct roadway edge between node "N102" and node "N105" in the mine geographic information model, and node "N104" needs to be passed through. Therefore, the logically inferred path does not match the actual roadway connectivity, and the path strength value of path A is multiplied by a connectivity penalty coefficient less than 1. The node sequence "N201 -> N203" of path B is directly connected in the actual roadway model, and path B passes the connectivity verification and does not need to be penalized in strength. A time decay factor is introduced. Path A is generated based on data five minutes ago, and path B is generated based on current data. For path A, its path strength value needs to decay over time according to a preset decay rate. Path B does not undergo time decay.

[0101] In some embodiments, all initial risk deduction paths that pass the spatio-temporal verification are classified according to the type of final disaster event they point to. Path A points to "gas explosion", path C also points to "gas explosion", path B points to "coal spontaneous combustion", and path D also points to "coal spontaneous combustion". The path strength value of path A is 0.65, the path strength value of path C is 0.50, the path strength value of path B is 0.80, and the path strength value of path D is 0.40. The probability fusion calculation is performed on all path strength values pointing to the same disaster event type to generate a comprehensive risk probability value. The fusion calculation uses the formula

[0102] ;

[0103] wherein: represents the comprehensive risk probability value of disaster event type t, represents the set of all initial risk deduction paths pointing to disaster event type t, represents the path strength value of the pth path in set G. It can be understood that this formula is based on the complementary principle of independent probability. Path A and path C point to "gas explosion", and their comprehensive risk probability value is calculated as Path B and path D point to "coal spontaneous combustion", and their comprehensive risk probability value is calculated as , the comprehensive risk deduction conclusion and the corresponding disaster event description constitute a comprehensive risk deduction conclusion, refer to Table 1.

[0104] Table 1: List of comprehensive risk deduction conclusions

[0105] Comprehensive risk deduction conclusion number Disaster event type Comprehensive risk probability value Main roadway nodes involved Conclusion 1 Gas explosion 0.825 N101, N102, N104, N105 Conclusion 2 Coal spontaneous combustion 0.88 N201, N203

[0106] In a specific implementation, the early warning generation module formats and generates a graded early warning instruction according to the emergency level and the influence range of the comprehensive risk deduction conclusion. The module internally pre-sets a risk probability threshold and an emergency level table for different categories of disaster events. For example, the risk probability threshold for the "gas explosion" disaster event is set as: [0, 0.6) corresponds to the "attention" level, [0.6, 0.8) corresponds to the "warning" level, and [0.8, 1.0] corresponds to the "alert" level. The "coal spontaneous combustion" disaster event sets different threshold intervals. The comprehensive risk probability value in the comprehensive risk deduction conclusion is compared with the risk probability threshold. The comprehensive risk probability value 0.825 of conclusion 1 belongs to the interval "[0.8, 1.0]", and the emergency level to which it belongs is determined to be "alert". The comprehensive risk probability value 0.88 of conclusion 2 belongs to its corresponding "alert" interval.

[0107] In some embodiments, the influence range level of the comprehensive risk deduction conclusion is determined according to the range of the affected roadway nodes. Conclusion 1 involves nodes "N101", "N102", "N104", and "N105", which are distributed in two mining areas. According to the preset rule, the influence range level is determined to be "cross-mining area". Conclusion 2 involves nodes "N201" and "N203", which are located in the same working face. The influence range level is determined to be "within working face". Combining the emergency level and the influence range level, the corresponding instruction template is selected from the early warning instruction template library. Optionally, for the disaster type "gas explosion" with emergency level "alert" and influence range "cross-mining area", the instruction template with template number "Alert_CrossArea_GasExplosion" is selected. For the disaster type "coal spontaneous combustion" with emergency level "alert" and influence range "within working face", the instruction template with template number "Alert_WorkingFace_SpontaneousCombustion" is selected. The specific description of the comprehensive risk deduction conclusion and the involved roadway location information are filled into the selected instruction template to generate the final graded early warning instruction.

[0108] The construction steps of the mine geographic information model include: obtaining point cloud data of the mine tunnel by a three-dimensional laser scanner, and synchronously importing design drawings and exploration data of the tunnel, denoising and registering the point cloud data, and generating an accurate three-dimensional spatial point cloud model of the tunnel. Based on the three-dimensional spatial point cloud model of the tunnel, a spatial clustering and feature recognition algorithm is used to segment and extract point cloud sets of each independent tunnel, and then the center axis and cross section profile of each tunnel are fitted. The intersection of the center axis of each tunnel at both ends and other axes is defined as a node, and the center axis segment connecting two nodes is defined as an edge, so as to abstract the tunnel network into a graph structure composed of nodes and edges. The three-dimensional coordinates of each node in the graph are calculated, and the length, slope, azimuth and connected upstream and downstream node identifiers of each edge are calculated, and these attributes are attached to the corresponding graph elements. The tunnel topology graph with attributes is associated and encapsulated with the three-dimensional spatial point cloud model of the tunnel to form a mine geographic information model containing geometric model and topological relationship. The construction steps of the mine disaster knowledge graph include: collecting the associated knowledge between gas anomalies, geological structure, mining activities and various mine disaster events from mine geological reports, safety procedure documents, historical accident case library and field expert experience manual. The ontology structure of the knowledge graph is defined, the core entity types and their attributes are determined, the core entity types include geological entities, gas states, environmental parameters, human activities and disaster events; the relationship types between entities are defined, the relationship types include causal relationship, accompanying relationship, triggering relationship and inhibition relationship. For structured historical case data, directly map entity attributes and relationships, and extract "reason-event-result" triples. For unstructured text reports and procedures, use natural language processing techniques for named entity recognition and relationship extraction, and extract graph substructures with entities as nodes and relationships as edges. The entities, relationships and attributes extracted from all sources are aligned, disambiguated and fused under a unified ontology framework to construct a mine disaster knowledge graph.

[0109] In a specific implementation, taking a coal mine with an annual output of 1.2 million tons as an example, the mine geographic information model is constructed. The main transport roadway, return airway and coal mining face roadway of the east wing mining area of the mine are scanned by RIEGL VZ-400i type three-dimensional laser scanner to obtain point cloud data of the roadway surface, and the point cloud density reaches 5000 points per square meter. The CAD design drawings and geological exploration drilling data of the mining area are imported synchronously, the original point cloud data obtained is denoised to remove outliers caused by equipment, pipelines and personnel interference, and the point cloud data obtained by multiple station scanning is registered through target points and unified to the mine absolute coordinate system to generate an accurate three-dimensional space point cloud model of the roadway. Based on the three-dimensional space point cloud model of the roadway, the DBSCAN spatial clustering algorithm is used to segment and extract the point cloud sets of each independent roadway of the "east wing transport roadway" and "1101 working face return airway" according to the spatial density and continuity of the point cloud. The specific implementation of the DBSCAN spatial clustering algorithm based on the three-dimensional space point cloud model of the roadway is as follows: first, the neighborhood radius parameter is set according to the approximate physical size of the roadway, and the minimum point threshold required to form a core object is determined based on the average density of the scanned point cloud. The algorithm starts from any point in the point cloud, calculates the number of points in its neighborhood, and if the threshold is reached, the point is marked as a core point and a new cluster is created. Subsequently, all points that can reach the core point density are iteratively assigned to the same cluster. This process expands according to the local density difference and spatial continuity of the point cloud in three-dimensional space, and can identify the continuous and dense roadway point cloud region in space as an independent cluster, thereby segmenting the point cloud set of each independent roadway such as the "east wing transport roadway". Sparse points that cannot be assigned to any cluster are identified as noise. The principal component analysis method is used to fit the main direction of the point cloud set of each roadway, and then the center axis of each roadway is obtained. Point cloud slices are taken along the center axis at certain intervals, and the least squares method is used to fit the rectangular cross section to obtain the cross section profile parameters. For example, the "1101 working face return airway" is fitted as a rectangular profile with a width of 5 meters and a height of 3.5 meters.

[0110] In some embodiments, the intersection of the center axis of each roadway and its other axis is defined as a node in the graph structure, for example, the intersection of the "east wing transport roadway" and the "1101 working face return airway" is defined as node "N101", and the center axis segment connecting two nodes is defined as an edge, for example, the roadway segment connecting node "N101" and node "N102" is defined as edge "E001", thereby abstracting the roadway network into a graph structure composed of nodes and edges. The three-dimensional coordinates of each node in the graph in the mine absolute coordinate system are calculated, for example, the coordinates of node "N101" are (X:12543.21, Y:5678.34, Z:-455.30), and the length, slope, azimuth angle and connected upstream and downstream node identifiers of each edge are calculated, and the three-dimensional coordinates, length, slope, azimuth angle and node identifiers are attached to the corresponding graph elements as attributes, for example, the attributes of edge "E001" are recorded as {length: 150.5 meters, slope: 3°, azimuth angle: 85°, upstream node: N101, downstream node: N102}. The roadway topology graph with attributes is associated and encapsulated with the roadway three-dimensional space point cloud model, and the nodes and edges in the topology graph are linked to the corresponding space regions in the point cloud model through a unified index file, forming a mine geographic information model containing both geometric model and topological relationship, which is stored in a standard file format for calling by the spatial association module.

[0111] In specific implementation, the construction steps of the mine disaster knowledge graph start from multi-source knowledge collection, including "mine east wing geological structure specification" mine geological report, "coal mine safety regulations" document, enterprise internal nearly ten years of historical accident case library and "gas and fire prevention experience manual" compiled by three senior safety engineers. These data cover the associated knowledge between gas anomaly, fault structure, mining activity and gas explosion, coal spontaneous combustion and other mine disaster events. The ontology structure of the knowledge graph is defined, the core entity types and their attributes are determined, the core entity types include geological entities, gas state, environmental parameters, human activities and disaster events; the relationship types between entities are defined, including causal relationship, accompanying relationship, triggering relationship and inhibition relationship.

[0112] In some embodiments, for structured historical case data, such as a record in an accident database containing "Accident Type: Gas Explosion", "Direct Cause: Gas Accumulation in Goaf", and "Inducing Factor: Poor Ventilation in Blind Lane", entity attributes and relationships are directly mapped, and two "cause-event-effect" triples, "Gas Accumulation in Goaf - Leads to -> Gas Explosion" and "Poor Ventilation in Blind Lane - Promotes -> Gas Accumulation in Goaf", are extracted. For unstructured text reports and procedures, such as a description in a geological report, "The coal seam near the F101 fault zone is broken, and gas is prone to accumulate, making it a dangerous area for gas outbursts", named entity recognition is performed using natural language processing technology based on the BERT model. This identifies "F101 fault zone", "coal seam", "gas", and "gas outburst", and relationships are extracted to create a graph substructure with entities as nodes and relationships as edges, such as "F101 fault zone - Leads to -> Coal Seam Breakage" and "Coal Seam Breakage - Prone to Accumulation -> Gas". Entities, relationships, and attributes extracted from all sources are aligned, disambiguated, and merged within a unified ontology framework. For example, the entity "exceeding gas limits" extracted from procedures and the entity "high concentration of gas" extracted from cases are aligned to the same standard entity "abnormal gas concentration," constructing a mine disaster knowledge graph containing thousands of entities and tens of thousands of relationships. This graph is stored in a graph database for querying by the risk simulation module. The weights of relationships in the knowledge graph are also considered. The initial value can be assigned based on the authority of its source and the frequency of historical verification. The calculation formula is as follows:

[0113] ;

[0114] in: This represents the initial weight of the relationship. This indicates the source's authority and rating. This represents the normalized value of the frequency of the relationship in historical cases. It is a preset coefficient that balances the influence of both.

[0115] See Figure 5 This is a pie chart showing the distribution of mine roadway conditions. It corresponds to the statistical results of the "data collection and status analysis" phase in the mine gas parameter early warning system, primarily used to display the percentage of each roadway condition. The roadways are predominantly in normal condition, accounting for over 80%; abnormal conditions account for 16.8% in total, with "fire hazard" and "gas leakage" being the main risk types (totaling 14.8%), while "poor ventilation" has a lower percentage but is a potential contributing factor. This type of chart is a macro-level statistical tool for mine safety status—by quantifying the percentage of each condition, the overall safety level of the roadways can be quickly grasped, providing a basic basis for subsequent "risk simulation" and "early warning generation," and helping to prioritize high-percentage risk types.

[0116] It has to be noted that, in the present document, the terms "first", "second", etc. merely serve to identify a subject or action, without necessarily requiring or implying any such actual relationship or order between such subjects or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0117] While embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, combinations, and variations of the embodiments can be undertaken without departing from the spirit and scope of the present application, which is defined by the appended claims and their equivalents.

Claims

1. A mine digital gas parameter early warning system based on gas analysis, characterized in that, The method comprises the following steps: a data acquisition and feature extraction module acquires original gas data streams and auxiliary environmental data streams generated by various sensors distributed in the mine roadway, and extracts features of the original gas data streams and auxiliary environmental data streams to form a gas state behavior pattern; a space correlation module correlates the gas state behavior pattern with the roadway space topology according to a mine geographic information model to generate a gas state behavior pattern carrying a space label; an intention analysis module is used to match and decompose the gas state behavior pattern carrying a space label based on a pre-defined gas state intention type library to generate a plurality of basic gas state intention segments for a specific roadway, and assign a confidence score to each basic gas state intention segment to construct an intention segment set; a risk deduction module calls a pre-defined mine disaster knowledge graph to find a matching correlation rule for each basic gas state intention segment in the intention segment set, activate the associated graph nodes and edges to form an initial risk deduction path, and perform spatio-temporal consistency verification and probability integration on the initial risk deduction path to generate one or more comprehensive risk deduction conclusions; a warning generation module formats a graded warning instruction according to the emergency degree and influence range of the comprehensive risk deduction conclusion; the step of performing spatio-temporal consistency verification and probability integration on the initial risk deduction path comprises: obtaining the spatial position information of all roadway nodes involved in the initial risk deduction path; checking the connectivity of adjacent nodes in the path in the mine geographic information model, and if the logical deduction path does not match the actual roadway connectivity, reducing the path strength value of the initial risk deduction path; introducing a time decay factor, and for the initial risk deduction path generated based on earlier time data, the path strength value decays over time; classifying all initial risk deduction paths that pass the spatio-temporal verification according to the final disaster event type pointed by the path; performing probability fusion calculation on the path strength values of all paths pointing to the same disaster event type to generate a comprehensive risk probability value, and the comprehensive risk probability value and its corresponding disaster event description jointly constitute a comprehensive risk deduction conclusion; the step of formatting a graded warning instruction comprises: pre-setting a risk probability threshold and an emergency degree level table for different categories of disaster events; comparing the comprehensive risk probability value in the comprehensive risk deduction conclusion with the risk probability threshold to determine its emergency degree level; determining the influence range level according to the range of the roadway nodes affected by the comprehensive risk deduction conclusion; combining the emergency degree level and the influence range level to select the corresponding instruction template from the warning instruction template library; filling the specific description of the comprehensive risk deduction conclusion and the involved roadway position information into the selected instruction template to generate the final graded warning instruction.

2. A digital gas parameter warning system for mine based on gas analysis according to claim 1, characterized in that, the step of extracting features of the original gas data streams and auxiliary environmental data streams to form a gas state behavior pattern comprises: synchronizing and time-aligning the original gas data streams and auxiliary environmental data streams to generate a standardized fusion data sequence; Extracting gas concentration rate of change, environmental parameter covariant trend and historical abnormal segment recurrence feature from the fusion data sequence to form a gas state behavior pattern; the specific steps include: For the gas concentration data in the fusion data sequence, a sliding time window is used to calculate the change gradient in unit time, and the change gradient is recorded as the gas concentration rate of change; Synchronously analyzing the temperature and humidity data in the fusion data sequence, calculating the covariance matrix of the gas concentration rate of change, and determining the environmental parameter covariant trend according to the principal component direction of the covariance matrix; Retrieving the historical gas database, calling the historical data with similar spatial tags as the current fusion data sequence, comparing the current sequence with the historical abnormal segment by using the dynamic time warping algorithm, calculating the similarity, and marking the similarity results exceeding the threshold as the historical abnormal segment recurrence feature; The gas concentration rate of change, environmental parameter covariant trend and historical abnormal segment recurrence feature are normalized and spliced to form a multi-dimensional feature vector, which is the gas state behavior pattern.

3. The digital gas parameter early warning system for mine based on gas analysis according to claim 1, characterized in that, The step of associating the gas state behavior pattern with the roadway space topology according to the mine geographic information model includes: Reading the mine geographic information model, which defines the roadway nodes and connection edges in a graph structure; Mapping the physical location of the sensor corresponding to the gas state behavior pattern to the nearest roadway node in the mine geographic information model; According to the mapping relationship, the unique identifier of the roadway node and the identifier list of adjacent nodes are attached to the gas state behavior pattern to generate a gas state behavior pattern carrying a spatial tag; The gas state behavior pattern carrying a spatial tag not only contains gas features, but also implies its propagation and influence potential in the roadway network.

4. The digital gas parameter early warning system for mine based on gas analysis according to claim 1, characterized in that, The step of intent matching and decomposition of the gas state behavior pattern carrying a spatial tag based on a pre-defined gas state intent type library includes: The gas state intent type library enumerates a plurality of independent gas abnormal states, each of which is defined by a set of feature condition rules; The multi-dimensional feature vector in the gas state behavior pattern carrying a spatial tag is compared with the feature condition rules of each gas abnormal state in the gas state intent type library one by one; When the multi-dimensional feature vector meets part or all of the feature condition rules of a certain gas abnormal state, it is determined that the gas state behavior pattern carrying a spatial tag contains the gas abnormal state, and a basic gas state intent segment is generated; One gas state behavior pattern carrying a spatial tag matches and decomposes multiple basic gas state intent segments, and each basic gas state intent segment represents a potential abnormal type identified.

5. A digital gas parameter warning system for mine based on gas analysis according to claim 4, characterized in that, The step of assigning a confidence score to each basic gas state intent segment to construct an intent segment set includes: For each generated basic gas state intent segment, retrieve the feature condition rules on which it is matched; Counting the number of items in the multi-dimensional feature vector that satisfy the feature condition rule, and calculating the proportion of the number of items satisfying the total number of rule items as the rule satisfaction degree; Combining the similarity value of the historical abnormal segment reproduction characteristics corresponding to the basic gas state intention picture segment, the rule satisfaction degree is weighted and corrected to obtain an initial confidence degree; According to the activity degree of the roadway node where the basic gas state intention picture segment is located in the historical disaster, the initial confidence degree is adjusted twice to generate a final confidence score; All basic gas state intention picture segments and their corresponding confidence scores together constitute an intention picture segment set.

6. The digital gas parameter early warning system for mine based on gas analysis according to claim 1, characterized in that, The step of calling a predefined mine disaster knowledge graph to find a matching associated rule for each basic gas state intention picture segment in the intention picture segment set includes: Parsing the mine disaster knowledge graph, wherein the nodes in the mine disaster knowledge graph represent geological entities, gas states, or disaster events, and the edges represent causal relationships or accompanying relationships; Taking the basic gas state intention picture segment as a query input, find the node with the basic gas state intention picture segment name as the core in the mine disaster knowledge graph; Starting from the found core node, traverse all directly connected edges and adjacent nodes, which constitute the associated rule directly related to the basic gas state intention picture segment; Record the activated nodes and edges as the local reasoning subgraph of the basic gas state intention picture segment.

7. A mine digital gas parameter early warning system based on gas analysis according to claim 6, characterized in that, The step of activating related graph nodes and edges to form an initial risk deduction path includes: For all basic gas state intention picture segments in the intention picture segment set, superimpose their respective activated local reasoning subgraphs; In the superimposed graph, identify the nodes and edges that are activated by multiple local reasoning subgraphs, which become high-risk convergence points; Taking the high-risk convergence point as the core, backtrack and connect to each basic gas state intention picture segment node as the starting point of the query to form one or more reasoning chains from intention picture segments to potential disaster events, which are the initial risk deduction paths; Each of the initial risk deduction paths is attached to a path strength value, which is calculated from the confidence scores of all basic gas state intention picture segments and the weights of the edges passed through.

8. The digital gas parameter warning system for mine based on gas analysis according to claim 1, characterized in that, The construction steps of the mine geographic information model include: Obtain point cloud data of the mine roadway by a three-dimensional laser scanner, and simultaneously import the design drawings and exploration data of the roadway, perform denoising and registration processing on the point cloud data to generate an accurate three-dimensional spatial point cloud model of the roadway; Based on the three-dimensional spatial point cloud model of the roadway, use spatial clustering and feature recognition algorithms to segment and extract the point cloud set of each independent roadway, and then fit the center axis and cross section profile of each roadway; Define the intersection of the center axis of each roadway and other axes as a node, and define the center axis segment connecting two nodes as an edge, thereby abstracting the roadway network into a graph structure composed of nodes and edges; Calculate the three-dimensional coordinates of each node in the graph, and calculate the length, slope, azimuth and connected upstream and downstream node identifiers of each edge, and attach the three-dimensional coordinates, length, slope, azimuth and node identifiers as attributes to the corresponding graph elements; Correlate and encapsulate the roadway topology graph with attributes with the roadway three-dimensional space point cloud model to form a mine geographic information model containing both geometric models and topological relationships.

9. The digital gas parameter warning system for mine based on gas analysis according to claim 1, characterized in that, The construction steps of the mine disaster knowledge graph include: Collecting associated knowledge between gas anomalies, geological structures, mining activities and various mine disaster events from mine geological reports, safety procedure documents, historical accident case libraries and field expert experience manuals; Defining the ontology structure of the knowledge graph, determining the core entity types and their attributes, the core entity types including geological entities, gas states, environmental parameters, human activities and disaster events; defining the relationship types between entities, the relationship types including causal relationship, accompanying relationship, triggering relationship and inhibition relationship; For structured historical case data, directly map entity attributes and relationships, and extract "reason-event-result" triples; For unstructured text reports and procedures, use natural language processing techniques for named entity recognition and relationship extraction to extract graph substructures with entities as nodes and relationships as edges; Align, disambiguate and fuse the extracted entities, relationships and attributes from all sources under a unified ontology framework to construct a mine disaster knowledge graph.

Images