Coal mine emergency resource dynamic data intelligent analysis and scheduling system based on large model

Abstract

This invention relates to the field of coal mine safety production and emergency rescue technology, and discloses a coal mine emergency resource dynamic data intelligent analysis and scheduling system based on a large model. The system includes: a semantic causal knowledge base construction module for parsing and constructing an experience base for emergency plans; a dynamic physical map maintenance module for real-time updating of roadway access costs; an abnormal event vectorization module for generating environmental feature vectors; a virtual impedance map correction module for constructing an anisotropic emergency situation map by combining emergency plan logic and environmental features; a resource scheduling planning module for generating conflict-free time-series trajectories using a moving congestion window; and a scheduling instruction generation and interaction module for converting instructions and dynamically allocating communication resources. This invention achieves anisotropic path planning by mapping the semantic logic of the emergency plan to the virtual impedance of the physical topology, and utilizes a spatiotemporal congestion detection mechanism to resolve multi-machine collaboration conflicts in confined spaces, thereby improving the safety and real-time performance of resource scheduling in disaster environments.

Description

Technical Field

[0001] This invention relates to the field of coal mine safety production and emergency rescue technology, specifically a coal mine emergency resource dynamic data intelligent analysis and scheduling system based on a large model. Background Technology

[0002] The underground production environment in coal mines is complex, and disasters are highly unpredictable. Rapid and accurate dispatch of emergency rescue resources is crucial for ensuring personnel safety and minimizing property damage. Existing coal mine emergency command systems typically rely on Geographic Information Systems (GIS) and safety monitoring systems. These systems collect environmental data in real time through sensors deployed in the tunnels and utilize classical graph theory algorithms for path planning to manage the dispatch of rescue equipment and materials. These systems play a vital role in routine production scheduling, providing fundamental data support and decision-making assistance for underground logistics and personnel positioning, and are an important component of current coal mine informatization efforts.

[0003] In dealing with complex disaster environments, existing scheduling methods struggle to effectively integrate unstructured emergency plans with the dynamically changing underground physical environment. Traditional path planning algorithms often construct topology networks based on static physical distances or single environmental parameters, lacking a deep understanding and application of the expert experience and logical rules inherent in textual emergency plans. Especially in disaster scenarios such as gas outbursts or fires, disaster evolution exhibits clear directional characteristics; for example, toxic gases spread with airflow. Conventional isotropic algorithms struggle to dynamically represent these risk gradient differences that change with airflow direction within the topology, resulting in planned paths that, while physically shortest, may lie in high-risk areas of disaster spread, failing to meet the tactical requirements of safety procedures regarding upwind rescue or avoiding spread paths.

[0004] Underground tunnels are spatially confined, posing physical challenges to the coordinated movement of large rescue equipment in narrow spaces. Existing scheduling models typically simplify mobile equipment as point masses, neglecting the spatial constraints between the equipment's geometry and the tunnel cross-section. This makes it difficult to predict the risk of congestion when multiple pieces of equipment converge during the planning phase, easily leading to physical congestion at critical nodes and disrupting rescue routes. Furthermore, scheduling instructions are often issued primarily using raw coordinate data or simple text information, lacking semantic conversion that aligns with human cognitive habits. In the low-bandwidth communication environment underground, the lack of dynamic transmission mechanisms based on the urgency of instructions can cause transmission delays for critical instructions involving changes to critical paths or warnings of high-risk areas, affecting the real-time performance and accuracy of on-site execution. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a dynamic data intelligent analysis and scheduling system for coal mine emergency resources based on a large model. This system solves the problems of physical environment constraints and semantic logic separation of contingency plans in traditional coal mine emergency scheduling, insufficient quantification of the impact of dynamic disaster evolution on path planning, and conflicts in the coordination of multiple equipment in narrow spaces.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a dynamic data intelligent analysis and scheduling system for coal mine emergency resources based on a large model, comprising: The semantic causal knowledge base construction module is configured to parse emergency plans to extract disaster causal logic and resource adaptation rules, and build a digital expert experience base. The dynamic physical map maintenance module is configured to maintain a topological network that reflects the underground spatial relationships and update the basic physical access cost of roadway side sections in real time. The abnormal event vectorization module is configured to encode real-time monitoring data into environmental severity features and feature query vectors. The virtual impedance spectrum correction module is configured to use the contingency plan logic retrieved from the digital expert experience database by the feature query vector, and combine it with the environmental severity characteristics to perform directional sensitivity weighting on the topology network of the dynamic physical spectrum maintenance module, and construct an anisotropic emergency situation spectrum. The resource scheduling and planning module is configured to calculate a global scheduling scheme based on the anisotropic emergency situation map, perform conflict detection using a moving blocking window, and generate a collaborative time-series trajectory. The scheduling instruction generation and interaction module is configured to convert the global scheduling scheme generated by the resource scheduling planning module into natural language instructions, and dynamically allocate communication resources according to the transmission priority coefficient.

[0007] The semantic causal knowledge base construction module performs a structured transformation of unstructured text. It uses a named entity recognition model based on sequence labeling to process the preprocessed text sequence, outputting a set of entities with preset category labels. These labels cover disaster sources, affected areas, response actions, and resources and equipment. The module iterates through the entity set, combining disaster source entities and affected area entities within the same syntactic structure. It then uses a relational classification model to determine logical relationships and generate structured knowledge triples. Furthermore, the module parses the physical constraints between resources and the environment from the contingency plan text to construct a resource-environment adaptation matrix. This matrix is ​​configured as a two-dimensional array, using the unique identifier of resource and equipment entities as row indices and environmental severity indicators such as temperature, water depth, and gas concentration as column indices. Cell values ​​store the corresponding tolerance thresholds.

[0008] The dynamic physical map maintenance module constructs an undirected weighted graph consisting of a node set and an edge set, and calculates the basic physical passage cost based on the physical length of the roadway, the rated speed of the standard transport vehicle, and the average slope angle of the roadway edge. The module also connects to sensor data streams to monitor the roadway status in real time. When a physical blockage signal is detected, the basic physical passage cost of the corresponding edge is updated to an infinite value, and the path is marked as impassable in the topology network.

[0009] The abnormal event vectorization module transforms monitoring data into semantic and physical features. It uses a pre-defined event description template to convert the extracted time-series data into natural language event context text. This template includes the alarm location name, abnormal parameter type, degree of numerical exceedance, and evolution trend characteristics. Subsequently, a pre-trained deep neural network embedding model is used to vectorize the event context text, generating a feature query vector. Simultaneously, the module calculates a multi-dimensional environmental severity vector, encompassing thermal hazard, toxicity, line-of-sight, and deflagration severity components. Based on real-time sensor values, safety thresholds, and extreme disaster thresholds, the module uses a piecewise normalized mapping function to calculate the component values ​​for each dimension.

[0010] The virtual impedance map correction module, based on semantic reasoning and real-time environmental conditions, decomposes the disaster-affected roadway segments in the physical map into two directed edges with opposite directions, and calculates the dynamic correction cost of these directed edges. The calculation parameters for the dynamic correction cost include the basic physical passage cost, environmental severity index, and directional risk coupling factor. The virtual impedance map correction module utilizes the dot product relationship between the moving direction vector and the real-time wind direction vector, combined with the definition of disaster spread trend in the contingency plan, to calculate the directional risk coupling factor. When the contingency plan logic determines that the disaster spreads with the wind and the planned path direction is consistent with the wind direction, the value of the directional risk coupling factor increases the dynamic correction cost, thereby suppressing the selection of downwind paths, preserving the feasibility of upwind rescue channels, and generating an anisotropic emergency situation map.

[0011] The resource scheduling and planning module utilizes a moving congestion window mechanism to address the multi-equipment coordination problem within confined spaces. The moving congestion window includes a unique identifier for the current roadway segment, the expected entry time, the expected exit time, and the dynamic width of the roadway cross-section occupied by the equipment. The module compares the moving congestion window generated by the new task with the global spatiotemporal occupancy table to determine if there are geometrical traffic conflicts. A traffic conflict is determined when, within any overlapping time segment, the sum of the width of the newly added equipment and the widths of all existing equipment in the same roadway exceeds the effective passage width of the current roadway segment. If a congestion conflict is determined, the module calculates the avoidance waiting cost required to eliminate the conflict. This cost is determined based on the difference between the time when existing equipment in the conflicting equipment set completely exits the roadway segment and the initial time when the new equipment is expected to arrive at the roadway entrance, and is added to the total passage cost of the path.

[0012] The scheduling instruction generation and interaction module performs reverse geocoding, converting node indices into lane location names. It then uses a state machine-based slot-filling mechanism to automatically fill in the action type, target location, deadline, and constraints in the instruction template, generating natural language instructions. The module also calculates the transmission priority coefficient for each instruction in the queue to be sent. The calculation logic for the transmission priority coefficient is as follows: it is negatively correlated with the difference between the latest execution deadline of the task action corresponding to the scheduling instruction and the current time; positively correlated with the average environmental severity of the area involved in the scheduling instruction; and a weight value is added when the path changes, ensuring that critical instructions occupy the transmission slot first.

[0013] This invention provides an intelligent analysis and scheduling system for dynamic data of emergency resources in coal mines based on a large-scale model. It has the following beneficial effects: 1. This invention achieves a logical connection between unstructured contingency plan text and the underground physical topology network through a semantic causal knowledge base construction module and a virtual impedance spectrum correction module. The system transforms the extracted disaster evolution patterns into directional risk coupling factors in the anisotropic spectrum, dynamically correcting the edge weights in the planning algorithm. This mechanism enables the system to identify and avoid high-risk areas that spread with airflow when calculating paths, ensuring that the scheduling scheme complies with the constraints on airflow direction and disaster avoidance in coal mine safety regulations, and improving the adaptability of path planning to dynamic disaster environments.

[0014] 2. This invention utilizes a moving congestion window technique within a resource scheduling and planning module to achieve spatiotemporal coordinated control of multi-source heterogeneous equipment in confined spaces. The system employs a multi-dimensional constraint model incorporating time, spatial location, and dynamic occupancy width to detect geometrical traffic conflicts between devices in a global spatiotemporal occupancy table, and quantifies congestion risk based on avoidance and waiting costs. This method transforms the physical collision hazards in narrow tunnels into high-cost weights in path planning, guiding the algorithm to automatically generate collision-free trajectories with temporal intervals, ensuring the orderly passage of large rescue equipment in complex underground road networks.

[0015] 3. This invention achieves accurate conversion and efficient transmission of abstract scheduling data into standardized execution instructions through a scheduling instruction generation and interaction module. The system utilizes reverse geocoding and slot-filling mechanisms to generate natural language instructions with semantic information, and calculates transmission priority coefficients based on task deadlines and environmental severity to dynamically allocate communication channels. This technical solution ensures that critical control instructions involving high-risk areas or emergency path changes are prioritized in confined underground communication environments, guaranteeing the real-time performance and reliability of data interaction between the scheduling system and terminal equipment. Attached Figure Description

[0016] Figure 1 This is a structural block diagram of the intelligent analysis and scheduling system for dynamic data of emergency resources in coal mines based on a large model, according to the present invention. Figure 2 A schematic diagram illustrating the process of constructing the semantic causal knowledge base of this invention; Figure 3 This is a schematic diagram of the dynamic physical map maintenance process of the present invention; Figure 4 This is a schematic diagram illustrating the process of abnormal event vectorization and environmental severity feature encoding of the present invention; Figure 5 This is a schematic diagram of the virtual impedance spectrum correction and anisotropy weight calculation process of the present invention. Figure 6 This is a schematic diagram of the process for calculating the narrow space movement blocking window and coordinating multi-resource scheduling according to the present invention. Figure 7 This is a schematic diagram of the standardized scheduling instruction generation and multimodal interaction process of the present invention.

[0017] Among them, 10. Semantic causal knowledge base construction module; 20. Dynamic physical map maintenance module; 30. Abnormal event vectorization module; 40. Virtual impedance map correction module; 50. Resource scheduling planning module; 60. Scheduling instruction generation and interaction module. Detailed Implementation

[0018] The technical solutions in 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.

[0019] See attached document Figure 1 , Figure 1 This is a structural block diagram of a coal mine emergency resource dynamic data intelligent analysis and scheduling system based on a large model, according to an embodiment of the present invention. The present invention provides a coal mine emergency resource dynamic data intelligent analysis and scheduling system based on a large model, comprising: Semantic causal knowledge base construction module 10 is used to parse emergency plans to extract disaster causal pairs and resource adaptation rules, and build a digital expert experience base containing unstructured text and structured logic. The dynamic physical map maintenance module 20 is used to maintain the topology network reflecting the underground spatial relationships and airflow direction, access sensor data in real time to update the basic access cost, and mark the roadway airflow and size attributes. The abnormal event vectorization module 30 is used to encode abnormal monitoring data into environmental severity features in real time, generate event text and query vectors describing the disaster situation, and identify the threat level of the environment to resources; The virtual impedance spectrum correction module 40 is used to perform directional sensitivity weighting on the spectrum based on the contingency plan logic, and to construct an asymmetric virtual impedance by combining the wind direction and the contingency plan prediction to form an anisotropic emergency situation spectrum. The resource scheduling and planning module 50 is used to calculate the global optimal solution for spatiotemporal collision avoidance, select suitable resources, and generate a moving blocking window using a dynamic occupancy projection mechanism to achieve conflict-free and efficient passage of multiple vehicles. The scheduling instruction generation and interaction module 60 is used to generate standardized interaction instructions and use a large model to transform path data and timing logic into natural language scheduling commands that conform to coal mine specifications.

[0020] The specific implementation methods of each module of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] See attached document Figure 2 - Appendix Figure 3 , Figure 2 This is a schematic diagram illustrating the process of constructing a semantic causal knowledge base according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the process for maintaining a dynamic physical map according to an embodiment of the present invention.

[0022] In the intelligent analysis and scheduling system for dynamic data of emergency resources in coal mines based on a large model, the semantic causal knowledge base construction module 10 is responsible for establishing a hybrid data index containing unstructured emergency plan text, structured causal logic, and resource supply and demand matching rules. The semantic causal knowledge base construction module 10 reads existing historical emergency plan documents, accident analysis reports, and safety procedure documents from the coal mine through a data interface, and performs word segmentation and stop word removal preprocessing on the read text data. For the preprocessed text sequence, the semantic causal knowledge base construction module 10 uses a named entity recognition model based on sequence labeling architecture for processing. The input data of the named entity recognition model is configured as a natural language text sequence of the coal mine emergency plan, and the output data is configured as a set of entities with preset category labels. The preset category labels specifically include: entities describing the origin of the accident (such as gas outburst points), entities describing the affected area (such as return airways and refuge chambers), entities describing the response measures (such as power outages and sprinklers), and entities describing the resource equipment (such as powerful drainage pumps).

[0023] After obtaining the entity set, the semantic causal knowledge base construction module 10 performs relation classification operations. Specifically, the semantic causal knowledge base construction module 10 first traverses the identified entity set, pairing disaster source entities and disaster-affected area entities that are in the same natural paragraph or the same syntactic structure to form candidate entity pairs. Subsequently, the semantic causal knowledge base construction module 10 uses a pre-trained relation classification model to calculate the contextual semantic features of the candidate entity pairs to determine whether there is a logical association between the entity pairs that causes or triggers the relationship. When the calculated association confidence exceeds a preset threshold, the semantic causal knowledge base construction module 10 fills the corresponding disaster source entity into the subject position of the triple and the disaster-affected area entity into the object position of the triple, thereby generating a structured knowledge triple that conforms to the format of <disaster triggering condition, causes, affected associated area>.

[0024] In addition to causal logic, the semantic causal knowledge base construction module 10 parses the physical constraints between resources and the environment from the plan text and constructs a resource-environment adaptation matrix. The data structure of the resource-environment adaptation matrix is ​​configured as a two-dimensional array: using the unique identifier of the resource equipment entity as the row index, and using environmental severity indicators (including at least the upper limit of temperature, water depth, and gas concentration) as the column index, with the cell values ​​storing the corresponding tolerance thresholds. The semantic causal knowledge base construction module 10 stores the above structured triples, the resource-environment adaptation matrix, and the high-dimensional semantic vector representation of the original plan text in a hybrid storage architecture of graph database and vector database, forming a digital expert experience base that includes logical reasoning and semantic retrieval functions.

[0025] While constructing a digital expert experience database, the dynamic physical map maintenance module 20 is responsible for maintaining the topology network reflecting the spatial relationships of underground entities and the flow direction of the ventilation network. The dynamic physical map maintenance module 20 reads geographic information system data from the coal mine and constructs an undirected weighted graph consisting of node sets and edge sets. The node set data corresponds to the coordinates of underground roadway intersections, slope change points, and chamber entrances, while the edge set data corresponds to the roadway connections between two nodes. The dynamic physical map maintenance module 20 writes the physical length, average slope, road surface type, and cross-sectional geometric dimensions of the roadway into the edge attributes of the graph.

[0026] To quantify the passage capacity of the tunnel, the dynamic physical map maintenance module 20 calculates the basic physical passage cost for each segment. Considering the nonlinear hindering effect of the underground tunnel slope on the speed of transport vehicles, the dynamic physical map maintenance module 20 calculates the cost based on the following formula. The node and the first The basic physical passage cost of connecting edges between nodes : ; In the formula, It represents the basic physical cost of passage, which in physical terms is the time cost required for a standard vehicle to pass through a road segment; Indicates the first The node and the first The physical length of the tunnel between nodes; This indicates the rated speed of a standard transport vehicle in a horizontal tunnel. This represents the average slope angle of the roadway side section; This represents the slope resistance correction coefficient, which is preset as a fixed constant based on the friction characteristics of underground coal mine roads and vehicle dynamics.

[0027] The dynamic physical map maintenance module 20 accesses the sensor data stream of the underground safety monitoring system in real time via the industrial ring network. The module maps wind speed and direction sensor data distributed in different roadways to the corresponding edges of the topology network, updating the airflow direction status and wind speed values ​​of each roadway in real time in the map attributes. When roadway sensors detect physical obstruction signals, including sudden changes in roof separation meter data indicating collapse or water level sensor exceeding limits indicating flooding, the dynamic physical map maintenance module 20 updates the basic physical passage cost of the corresponding edge segment. The value is updated to infinity, thus marking the path as impassable at the topological network level. Through the above process, the dynamic physical map maintenance module 20 establishes a spatial data model that includes static geometric constraints, dynamic fluid properties, and real-time connectivity, providing an accurate physical reference system for subsequent superposition of virtual impedances and anisotropic path planning.

[0028] See attached document Figure 4 , Figure 4 This is a schematic diagram illustrating the process of vectorizing abnormal events and encoding environmental severity features according to an embodiment of the present invention.

[0029] In the complex monitoring environment of underground coal mines, the anomaly event vectorization module 30 performs transformation processing from raw sensor data to high-dimensional semantic features and physical severity features. The anomaly event vectorization module 30 continuously monitors the multi-source sensor data streams of the underground safety monitoring system through a data interface. The monitored indicators specifically include methane concentration, carbon monoxide concentration, ambient temperature, smoke concentration, and water level. When any monitored indicator... Exceeding the preset security threshold At that time, the exception event vectorization module 30 starts the exception data interception program to obtain the preset time window before and after the trigger time. Multidimensional sensor time-series data sequences within.

[0030] To transform discrete numerical fluctuations into semantic information understandable by a large model, the abnormal event vectorization module 30 uses a preset event description template to convert the extracted time-series data into event context text in natural language. The event description template is configured as a slot-filled structure containing the alarm location name, abnormal parameter type, degree of numerical exceedance, and evolution trend characteristics. For example, when gas data is abnormal, the generated event context text is: "Gas concentration (parameter) detected in the return airway (location) of the 1203 fully mechanized mining face is currently 1.5% (value), and has shown an exponential upward trend in the past 5-minute time window." After generating the event context text, the abnormal event vectorization module 30 calls a pre-trained deep neural network embedding model to vectorize the event context text. The input data of the deep neural network embedding model is configured as a character sequence of the event context text, and the output data is configured as a fixed-dimensional dense real-number vector, i.e., a feature query vector. Feature query vector It is configured to retrieve historical similar cases and associated contingency plans from the semantic causal knowledge base in subsequent steps.

[0031] In parallel with the semantic encoding process, to quantify the physical threat level of the current environment to different rescue equipment, the abnormal event vectorization module 30 calculates the environmental severity vector. The environmental severity vector is constructed as a multi-dimensional numerical vector, where each dimension represents the degree to which a specific environmental hazard hinders equipment operation or personnel activities. Based on real-time sensor values, the abnormal event vectorization module 30 uses a piecewise normalized mapping function to calculate the [missing information - likely a specific value or parameter] in the environmental severity vector. Component values ​​of each dimension The specific calculation formula is as follows: ; In the formula, The environmental severity vector represents the first... The severity scores are normalized values ​​for each dimension, ranging from [0, 1]. The larger the value, the more severe the environmental conditions. Indicates the first Real-time monitoring values ​​of environmental parameters; Indicates the first The safety critical threshold of environmental parameters is, in physical terms, the starting point of the restriction for normal equipment operation or normal personnel operation. Indicates the first The extreme disaster threshold of environmental parameters is, in physical terms, the value at which the equipment can no longer withstand the conditions or the environment reaches an extremely dangerous state.

[0032] The environment severity vector constructed by the abnormal event vectorization module 30 contains at least the following four dimensions of component values: Thermal damage severity component: Calculated based on temperature sensor data, reflecting the physical impact of high ambient temperature on the heat dissipation efficiency of electronic devices and battery discharge performance; Toxicity severity component: Calculated based on carbon monoxide and hydrogen sulfide sensor data, reflecting the degree of physiological harm to individuals or organisms without respiratory protection due to the concentration of toxic gases in the environment; Line-of-sight severity component: Calculated forward based on smoke sensor and dust concentration data, reflecting the degree of signal attenuation and interference of suspended particulate matter concentration in the environment on optical navigation devices and visual recognition systems; Deflagration severity component: Calculated based on gas and coal dust concentrations, it reflects the probability risk of a secondary explosion in the environment and is used to restrict access to non-explosion-proof equipment.

[0033] Through the above calculations, the abnormal event vectorization module 30 outputs two sets of key feature data: one set is a feature query vector used to retrieve contingency plan logic in the semantic space, and the other set is an environmental severity vector used to filter applicable resources in the physical space. These two sets of data transform heterogeneous sensor values ​​into decision features that the scheduling system can directly use, supporting the accurate screening and weight reduction of sensitive resources in subsequent stages.

[0034] See attached document Figure 5 , Figure 5 This is a schematic diagram of the process for virtual impedance spectrum correction and anisotropy weight calculation according to an embodiment of the present invention.

[0035] The virtual impedance map correction module 40, based on semantic reasoning results and real-time environmental conditions, adjusts the directional weights of the dynamic physical map, transforming the undirected physical topology network into an anisotropic emergency situation map with directional constraints. The virtual impedance map correction module 40 receives feature query vectors and environmental severity vectors output from the abnormal event vectorization module 30. Using the feature query vectors, the virtual impedance map correction module 40 searches the digital expert experience database generated by the semantic causal knowledge base construction module 10, recalling the contingency plan clauses with the highest matching degree to the current disaster scenario, and parsing the disaster spread logic and recommended evacuation and rescue directions under the current scenario. Specifically, when the retrieved disaster scenario is a fire in the intake airway, the virtual impedance map correction module 40 extracts from the contingency plan clauses the dynamic evolution logic of the disaster-derived smoke spreading downstream along the wind direction, as well as the tactical constraint rules defining the upwind direction as a safe rescue path.

[0036] Based on the retrieved logical rules, the virtual impedance map correction module 40 determines the set of potential nodes affected by the disaster in the map and performs a bidirectional decomposition operation on all connected edges within the potential node set. The virtual impedance map correction module 40 decomposes the roadway connected segments, originally represented as undirected edges in the physical map, into two directed edges with opposite directions, i.e., nodes. Pointing to node positive edge With nodes Pointing to node reverse edge After the decomposition is completed, the virtual impedance spectrum correction module 40 calculates the asymmetric dynamic virtual impedance by combining the real-time airflow direction data stored in the spectrum attributes.

[0037] To construct an anisotropic cost function that reflects the evolution of disasters with wind flow, and thus quantify the risk level differences of the same roadway in different travel directions (e.g., distinguishing between the disaster spread area on the downwind side and the safe entry area on the upwind side), the virtual impedance map correction module 40 introduces a directional risk coupling factor to differentiate the travel cost of each directed edge. The virtual impedance map correction module 40 calculates the directed edge according to the following formula. Dynamic correction cost : ; In the formula, Represents the corrected directed edge Dynamic passage cost; This represents the basic physical access cost provided by the dynamic physical map maintenance module 20; This represents the weighting coefficient of the contingency plan. The weighting coefficient is a preset positive real number used to adjust the amplification factor of environmental risks on the path weights. Representing the environmental severity index, the virtual impedance spectrum correction module 40 selects the component with the largest value in the current region's environmental severity vector as... The value is used to reflect the most significant hazard characteristics in the current area; This represents the directional risk coupling factor, used to quantify the correlation risk between the direction of movement and the direction of disaster evolution. The virtual impedance spectrum correction module 40 calculates based on the following logic. The value: ; in, Indicates from node Move to node The normalized unit direction vector is calculated from the difference in the spatial coordinates of the nodes; This represents the unit direction vector of airflow monitored in real time within the tunnel; This is a sign function that returns 1 when the vector dot product is positive and -1 when it is negative, representing the relationship between the direction of movement and the direction of airflow; This is a disaster spread trend identifier, and its value is determined according to the following rules: when the contingency plan logic determines that the disaster is spreading with the wind (such as fire smoke or gas spread), Set to 1; when the contingency plan logic determines that the disaster is spreading against the wind, Set to -1; when the contingency plan logic determines that the spread of disaster is unrelated to wind flow, Set to 0.

[0038] Through the calculations of the above formula, the virtual impedance spectrum correction module 40 achieves asymmetric assignment of values ​​for the same physical tunnel in different travel directions. Specifically, when the disaster spreads with the wind ( And when the planned path direction is consistent with the airflow direction (the dot product is positive), The calculation result is 1, making the dynamic passage cost... This significantly increases the impedance barrier at the algorithm level, inhibiting the path planning algorithm from selecting the direction of travel; conversely, if the planned path direction is opposite to the airflow direction, then... The calculation result is 0, and the dynamic passage cost falls back to the basic physical cost, thus preserving the feasibility of the headwind rescue channel.

[0039] The virtual impedance map correction module traverses all edges within the set of potential disaster-affected nodes, updates the weights of the entire graph, and generates an anisotropic emergency situation map that includes geometric distances, physical environmental constraints, and disaster evolution logic. This anisotropic emergency situation map serves as the data foundation for subsequent path planning, ensuring that the generated scheduling scheme is not only optimized in terms of distance but also logically compliant with the directional requirements for disaster avoidance route selection in the "Coal Mine Safety Regulations."

[0040] See attached document Figure 6 , Figure 6 This is a flowchart illustrating the calculation of the congestion window for movement in a narrow space and the collaborative scheduling of multiple resources according to an embodiment of the present invention.

[0041] In the confined spaces of underground coal mines, to address the potential physical congestion risks arising from the simultaneous operation of multiple rescue equipment in single-lane or restricted double-lane roadways, the spatiotemporal resource collaborative scheduling module 50 performs conflict detection and timing optimization based on a moving congestion window. The spatiotemporal resource collaborative scheduling module 50 receives a set of candidate paths generated by the virtual impedance spectrum correction module 40 and reads the physical dimension parameters of the equipment to be scheduled from the equipment database. These physical dimension parameters include at least the vehicle width, vehicle height, and minimum turning radius. Considering the non-uniformity and finiteness of the underground roadway cross-section, the spatiotemporal resource collaborative scheduling module 50 transforms traditional static path planning into dynamic spatiotemporal occupancy planning, constructing a multi-dimensional scheduling constraint model with timestamps, spatial location indices, and cross-sectional occupancy rates as dimensions.

[0042] To accurately characterize the dynamic occupancy of equipment within the roadway, the spatiotemporal resource coordination scheduling module 50 generates a movement blocking window for each piece of equipment en route or planned for deployment. The movement blocking window is configured as a data structure containing spatiotemporal constraint attributes, and its storage fields specifically include: a unique identifier for the roadway segment it resides in. Expected time of entering the side section of the roadway The estimated time of exiting the lane section. And the dynamic occupancy width of the roadway cross section by resource equipment during operation. The spatiotemporal resource collaborative scheduling module 50 maintains a global spatiotemporal occupancy table, which records the distribution data of the moving blocking windows of all roadway segments in the entire mine on the future time axis in real time through hash mapping or time slice rotation.

[0043] When the system plans a path for newly added rescue equipment, the spatiotemporal resource collaborative scheduling module 50 performs spatiotemporal collision detection on each lane segment of the candidate path. The module compares the movement blocking window generated by the new task with existing movement blocking windows in the global spatiotemporal occupancy table to determine if there is a geometrical traffic conflict. The traffic conflict determination logic is configured as follows: within any overlapping time segment, if the sum of the width of the newly added equipment and the width of all existing equipment in the same lane exceeds the effective passage width of the lane segment, a blocking conflict is determined to have occurred.

[0044] To resolve detected congestion conflicts, the spatiotemporal resource cooperative scheduling module 50 calculates the time compensation required to eliminate the conflict, i.e., the avoidance waiting cost. The spatiotemporal resource cooperative scheduling module 50 calculates the first... The resource equipment was attempting to enter the... The cost of waiting to avoid obstacles on the side of a lane : ; In the formula, This represents the minimum time delay required to eliminate the blockage, i.e., the time that new equipment needs to wait before entering the tunnel; Indicates the first The initial arrival time of each resource equipment at the tunnel entrance is estimated based on the current planned speed; This represents the set of conflicting equipment, which consists of all equipment in the global spacetime occupancy table within the time interval. Internal occupancy The existing equipment configuration of the side section of the alleyway; Indicates the first in the conflict equipment set The moment when all existing equipment has completely driven out of the side of the alley; This indicates a preset safety time interval buffer value to prevent rear-end collisions or scrapes; This is an indicator function. It takes the value of 1 when the inequality condition in parentheses is true (i.e., the sum of the widths of all equipment present is greater than the width of the roadway), and takes the value of 0 otherwise. Indicates the first The maximum physical width of each resource equipment; Indicates the first The physical width of an existing piece of equipment; Indicates the first The actual effective passage width of the side section of the alley after deducting the support structure and pipeline facilities.

[0045] Through the calculations using the above formulas, the spatiotemporal resource collaborative scheduling module 50 quantifies the additional time cost incurred by each potential path due to cross-sectional width constraints or queuing. The spatiotemporal resource collaborative scheduling module 50 then calculates the avoidance and waiting costs. The weight of the path is added to the total travel cost of the route, thereby achieving automatic weight reduction of high-congestion risk road sections at the scheduling algorithm level. For unsolvable conflict states that cannot be resolved by finite waiting (e.g., traveling in opposite directions in a one-way lane without a bypass chamber), if the characteristic function in the formula is true and the waiting time exceeds a preset threshold, the spatiotemporal resource collaborative scheduling module 50 marks the weight of the path as infinite, forcing the scheduling algorithm to replan the detour route.

[0046] After completing all conflict detection and cost correction, the spatiotemporal resource collaborative scheduling module 50 generates the final collision-free collaborative scheduling scheme. The collision-free collaborative scheduling scheme not only includes the spatial topology routes of each rescue equipment, but also precisely specifies the time thresholds (i.e., the earliest entry time and the latest exit time) of each equipment at key nodes. This ensures that in the narrow and complex underground environment, multi-source heterogeneous rescue forces can pass through bottleneck areas in an orderly manner in a time-sequential queue, thereby maximizing rescue efficiency.

[0047] See attached document Figure 7 , Figure 7 This is a schematic diagram of a standardized scheduling instruction generation and multimodal interaction process according to an embodiment of the present invention.

[0048] To convert the mathematical scheduling scheme generated by the spatiotemporal resource collaborative scheduling module 50 into action commands executable by frontline rescue personnel and control signals recognizable by equipment controllers, the scheduling command generation and interaction module 60 performs mapping processing from multi-dimensional spatiotemporal data to standardized communication protocols. The scheduling command generation and interaction module 60 receives a collision-free collaborative scheduling scheme containing a series of spatiotemporal trajectory points, a set of time thresholds, and avoidance action markers. Given that the collision-free collaborative scheduling scheme consists of abstract node indices and floating-point timestamps, the scheduling command generation and interaction module 60 calls the attribute database stored in the dynamic physical map maintenance module 20 to perform reverse geocoding. The scheduling command generation and interaction module 60 utilizes the node ID-to-lane name mapping table in the attribute database to convert the abstract node indices into specific lane location names, and adds the relative time offset to the current reference time to convert it into absolute clock time, thereby generating a task description sequence with semantic information.

[0049] The dispatch instruction generation and interaction module 60 employs a state machine-based slot-filling mechanism to generate natural language instructions, ensuring the standardization and consistency of the instruction content. The system has a pre-defined instruction template library, whose data structure is configured to include standardized fields such as action type, target location, deadline, and constraints. Based on the path attributes in the dispatch plan, the dispatch instruction generation and interaction module 60 automatically fills in the slots of the instruction templates. For example, when the dispatch plan instructs a vehicle to yield at a specific location, the dispatch instruction generation and interaction module 60 generates an instruction with the following content: Please proceed immediately to the 3rd refuge chamber (target location), arrive and park before 14:40 (deadline), and wait until the oncoming drainage vehicle passes (constraint).

[0050] To address the technical challenge of incompatible communication protocols among heterogeneous downhole equipment, the scheduling instruction generation and interaction module 60 encapsulates semantic instructions into control messages conforming to industry standards. The module 60 constructs a transmission data packet containing instruction metadata. The frame structure of the transmission data packet includes: a start identifier, a unique device identifier, an instruction sequence number, an encrypted instruction payload, a cyclic redundancy check (CRC) code, and an end identifier.

[0051] In low-bandwidth or high-latency communication environments underground, to ensure that critical safety instructions are transmitted with priority, the scheduling instruction generation and interaction module 60 calculates the transmission priority coefficient of each instruction in the instruction queue to be sent. The scheduling instruction generation and interaction module 60 calculates the priority coefficient of each instruction in the queue to be sent according to the following formula. Transmission priority of each scheduling instruction : ; In the formula, Indicates the first The transmission priority score of each instruction determines its order in the transmission buffer queue; the higher the score, the higher the priority. Indicates the latest deadline for the execution of the task action corresponding to the scheduling instruction; Indicates the current system time; This is a numerical stability factor used to avoid cases where the denominator is zero; This represents the average environmental severity of the area involved in the scheduling command, and its value is derived from the environmental severity vector calculated by the abnormal event vectorization module 30. This is a path change indicator factor. When the scheduling instruction is a change instruction caused by sudden congestion or environmental changes leading to path replanning, the value is 1; otherwise, the value is 0. , , These are the preset normalized coefficients for the time urgency weight, environmental risk weight, and change response weight, respectively, and they satisfy the following conditions: .

[0052] Through the above formula, the scheduling instruction generation and interaction module 60 realizes the dynamic allocation of communication resources, ensuring that critical instructions that are time-sensitive (close to the deadline), located in high-risk areas (high severity), or whose paths have changed can be given priority in being allocated time slots and sent to the terminal devices.

[0053] While distributing instructions, the dispatch instruction generation and interaction module 60 drives the human-machine interface terminal to display a visualized emergency situation map. The dispatch instruction generation and interaction module 60 reads the anisotropic spectrum data generated by the virtual impedance spectrum correction module 40, renders the planned path in a highlighted manner on the 3D digital twin interface, and overlays the environmental severity distribution as a heat map. For rescue vehicles involving manual driving, the dispatch instruction generation and interaction module 60 displays the generated natural language instructions on the vehicle terminal and sets up feedback controls for confirmation of receipt and inability to execute.

[0054] When the scheduling instruction generation and interaction module 60 receives a "cannot be executed" feedback signal, or fails to receive a response signal from the device within a preset timeout threshold, the scheduling instruction generation and interaction module 60 determines that the current scheduling instruction execution has failed. At this time, the scheduling instruction generation and interaction module 60 sends a road segment blockage marker to the dynamic physical map maintenance module 20 and triggers the spatiotemporal resource collaborative scheduling module 50 to start a local replanning program, thereby achieving closed-loop dynamic control of underlying parameters without manual intervention.

Claims

1. A coal mine emergency resource dynamic data intelligent analysis and scheduling system based on a large model, characterized in that, include: The semantic causal knowledge base construction module is configured to parse emergency plans to extract disaster causal logic and resource adaptation rules, and build a digital expert experience base. The dynamic physical map maintenance module is configured to maintain a topological network that reflects the underground spatial relationships and update the basic physical access cost of roadway side sections in real time. The abnormal event vectorization module is configured to encode real-time monitoring data into environmental severity features and feature query vectors. The virtual impedance spectrum correction module is configured to use the contingency plan logic retrieved from the digital expert experience database by the feature query vector, and combine it with the environmental severity characteristics to perform directional sensitivity weighting on the topology network of the dynamic physical spectrum maintenance module, and construct an anisotropic emergency situation spectrum. The resource scheduling and planning module is configured to calculate a global scheduling scheme based on the anisotropic emergency situation map, perform conflict detection using a moving blocking window, and generate a collaborative time-series trajectory. The scheduling instruction generation and interaction module is configured to convert the global scheduling scheme generated by the resource scheduling planning module into natural language instructions, and dynamically allocate communication resources according to the transmission priority coefficient.

2. The intelligent analysis and scheduling system for dynamic data of emergency resources in coal mines based on a large model as described in claim 1, characterized in that, The semantic causal knowledge base construction module uses a named entity recognition model based on sequence labeling architecture to process the preprocessed text sequence and outputs an entity set with preset category labels. The preset category labels include disaster source entities, disaster-stricken area entities, response action entities, and resource equipment entities. The semantic causal knowledge base construction module traverses the entity set, combines disaster source entities and disaster-stricken area entities that are in the same syntactic structure, uses a relation classification model to determine logical associations, and generates structured knowledge triples.

3. The intelligent analysis and scheduling system for dynamic data of emergency resources in coal mines based on a large model as described in claim 1, characterized in that, The resource environment adaptation matrix constructed by the semantic causal knowledge base construction module is configured as a two-dimensional array, with the unique identifier of the resource equipment entity as the row index and the environmental severity index as the column index. The cell values ​​of the matrix store the corresponding tolerance threshold. The environmental severity index includes at least the upper limit of temperature, the upper limit of water depth, and the upper limit of gas concentration.

4. The intelligent analysis and scheduling system for dynamic data of emergency resources in coal mines based on a large model as described in claim 1, characterized in that, The dynamic physical map maintenance module constructs an undirected weighted graph consisting of a node set and an edge set, and calculates the basic physical passage cost based on the physical length of the roadway, the rated speed of the standard transport vehicle, and the average slope angle of the roadway edge. When the roadway sensor detects a physical blockage signal, the dynamic physical map maintenance module updates the basic physical passage cost of the corresponding edge to an infinite value.

5. The intelligent analysis and scheduling system for dynamic data of emergency resources in coal mines based on a large model as described in claim 1, characterized in that, The abnormal event vectorization module uses a preset event description template to convert the captured time-series data into event context text in natural language form. The event description template is configured as a slot-filled structure that includes alarm location name, abnormal parameter type, degree of numerical exceedance, and evolution trend characteristics. The abnormal event vectorization module calls a pre-trained deep neural network embedding model to vectorize the event context text and generate a feature query vector for representing the disaster situation.

6. The intelligent analysis and scheduling system for dynamic data of emergency resources in coal mines based on a large model as described in claim 1, characterized in that, The environmental severity characteristics output by the abnormal event vectorization module are represented as a multi-dimensional numerical vector. The dimensions of the multi-dimensional numerical vector include thermal hazard severity components, toxicity severity components, line-of-sight severity components, and deflagration severity components. The abnormal event vectorization module calculates the component values ​​of each dimension in the multi-dimensional numerical vector based on real-time sensor values, safety critical thresholds, and extreme disaster thresholds using a piecewise normalized mapping function.

7. The intelligent analysis and scheduling system for dynamic data of emergency resources in coal mines based on a large model as described in claim 1, characterized in that, The virtual impedance spectrum correction module decomposes the roadway connection section affected by the disaster into two directed edges with opposite directions, and calculates the dynamic correction cost of the directed edges; the calculation parameters of the dynamic correction cost include the basic physical passage cost, environmental severity index and directional risk coupling factor. The directional risk coupling factor is used to quantify the associated risk between the direction of movement and the direction of disaster evolution. When the contingency plan logic determines that the disaster spreads with the wind flow and the planned path direction is consistent with the wind flow direction, the value of the directional risk coupling factor causes the dynamic correction cost to increase.

8. The intelligent analysis and scheduling system for dynamic data of emergency resources in coal mines based on a large model according to claim 1, characterized in that, The resource scheduling and planning module generates a movement blocking window for each resource piece of equipment, which includes a unique identifier of the roadway segment it belongs to, the expected entry time, the expected exit time, and the dynamic occupancy width of the roadway cross-section by the resource piece of equipment. The resource scheduling and planning module compares the movement blocking window generated by the new task with the global spatiotemporal occupancy table to determine whether there is a passage conflict in the geometric dimension. The condition for determining the passage conflict is: within any overlapping time segment, the sum of the width of the newly added equipment and the width of all existing equipment in the same roadway exceeds the effective passage width of the roadway segment it belongs to.

9. The intelligent analysis and scheduling system for dynamic data of emergency resources in coal mines based on a large model as described in claim 8, characterized in that, If a blockage conflict is determined to occur, the resource scheduling and planning module calculates the avoidance and waiting cost required to eliminate the conflict and adds it to the total passage cost of the path. The avoidance and waiting cost is determined based on the difference between the time when the existing equipment in the conflict equipment set completely leaves the roadway side and the initial time when the new equipment is expected to arrive at the roadway entrance. For unsolvable conflict states that cannot be eliminated by waiting, the resource scheduling and planning module marks the weight of the path as infinite and triggers detour route planning.

10. The intelligent analysis and scheduling system for dynamic data of emergency resources in coal mines based on a large model according to claim 1, characterized in that, The scheduling instruction generation and interaction module dynamically allocates communication resources based on the transmission priority coefficient. The calculated value of the transmission priority coefficient is negatively correlated with the difference between the latest execution deadline of the task action corresponding to the scheduling instruction and the current time, and positively correlated with the average environmental severity of the area involved in the scheduling instruction. Furthermore, the weight value is increased when the path changes.

Images