Robot group detection scheduling method and system

Abstract

The invention discloses a robot group detection scheduling method and system, and relates to disaster monitoring: scene reconstruction is carried out according to collected real-time sensing information, a three-dimensional digital model of a disaster scene is generated, and a discrete monitoring grid based on sampling points is established; constructing a disaster risk assessment model on the discrete monitoring points, and for gas and fire, calculating the risk value of each monitoring point by using actual measurement data to form a discrete risk distribution map; calculating the space change trend of the risk degree by adopting an adjacent point difference method, and calculating a directed gradient through the risk degree difference value and the distance of adjacent monitoring points; when the risk degrees of any two disasters are increased at the same time, marking the risk points as potential coupling risk points; identifying a propagation path of the risk degree based on connectivity analysis of the monitoring network; and scheduling the robot group by adopting a risk avoiding path planning algorithm. According to the method, the spatial distribution and evolution trend of the disaster field are accurately reconstructed through local gradient estimation under the sparse sampling condition.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of mine disaster monitoring, in particular to a robot group detection scheduling method and system. BACKGROUND

[0002] Mine disasters are a major hidden danger to the safety of coal mine production. Gas explosions, fires and other disasters often cause significant casualties and property losses. After a disaster occurs, quickly and accurately sensing the disaster site situation, predicting the disaster evolution trend, and developing a scientific rescue plan are the key to emergency rescue. Traditional manual detection methods pose a significant safety risk, and robot group collaborative detection has become an important technical means for mine disaster emergency response.

[0003] In the prior art, the development of mine disaster detection robot systems mainly focuses on the following aspects: real-time disaster information collection by robot groups, disaster scene reconstruction model construction, disaster scenario deduction algorithm design, disaster condition determination technology development, detection tactics and strategy library establishment, mine disaster detection task decision digital twin model construction, and cluster disaster collaborative detection command mechanism research. Related research institutions have developed system modules including perception information fusion, disaster identification, intelligent detection task decision, rapid generation of tactics and strategies, task assignment and autonomous scheduling, human-machine collaborative command, and command effectiveness evaluation, and have constructed a multi-disaster scene detection robot group scheduling command platform.

[0004] However, the prior art still has the following deficiencies in practical application: first, the robot group can only achieve sparse sampling in a complex mine environment, making it difficult to obtain complete disaster field information; second, there is insufficient understanding of the coupling evolution mechanism of multiple disasters such as gas and fire, and there is a lack of effective coupling risk identification methods; third, the disaster evolution trend prediction accuracy based on discrete sparse data is not high, and the spatial propagation characteristics of the disaster cannot be accurately grasped; fourth, the deployment and scheduling strategy of the robot group is relatively static and cannot be dynamically optimized according to the disaster evolution situation. These problems seriously restrict the implementation effect of rapid fusion and intelligent scheduling of detection information. SUMMARY

[0005] To address the problem of ineffective identification of multiple disaster coupling risks caused by sparse sampling of robot groups in a mine disaster environment in the prior art, the present application provides a robot group detection scheduling method and system that accurately reconstructs the spatial distribution and evolution trend of the disaster field under sparse sampling conditions through local gradient estimation.

[0006] One aspect of the present application provides a robot crowd detection scheduling method, comprising: S1, collecting real-time sensing information of a disaster site by a robot crowd, the real-time sensing information including environmental parameters, disaster feature data and spatial position data; wherein the environmental parameters include wind speed and direction data; S2, reconstructing a scene according to the collected real-time sensing information, generating a three-dimensional digital model of the disaster site, and establishing a discrete monitoring grid based on sampling points; S3, constructing a disaster risk assessment model on the discrete monitoring points, calculating the risk value of each monitoring point for gas and fire using the measured data, and forming a discrete risk distribution map; S4, calculating the spatial variation trend of the risk degree by using the difference method of adjacent points, and calculating the directed gradient through the risk difference and distance of adjacent monitoring points; when the risk degree of any two disasters increases at the same time, it is marked as a potential coupling risk point; S5, based on the connectivity analysis of the monitoring network, identifying the propagation path of the risk degree; S6, according to the risk distribution map and the directed gradient, using a risk avoidance path planning algorithm to schedule the robot crowd.

[0007] Further, S2, reconstructing a scene according to the collected real-time sensing information, generating a three-dimensional digital model of the disaster site, comprising: determining the absolute coordinates of each robot in the three-dimensional space based on the spatial position data of the robot crowd and the pre-stored mine tunnel topology structure; mapping the environmental parameters and disaster feature data collected by each robot to the corresponding spatial coordinate points to form a set of spatial sampling points with attribute information; according to the geometric characteristics of the mine tunnel, a three-dimensional grid framework is established along the tunnel direction and the vertical direction, and the grid nodes are set at the tunnel intersection points, turning points and equally spaced distribution points; the actual sampling points are mapped to the grid nodes, and for the grid nodes without sampling data, spatial interpolation is performed according to the data of adjacent nodes to obtain a discrete monitoring grid; a time series data cache is established for each grid node for time evolution analysis of the risk degree; an adjacency relationship table is established between the grid nodes according to the connectivity relationship of the tunnel, which is used for gradient calculation and connectivity analysis.

[0008] In particular, the key to accurately reconstructing the disaster site under sparse sampling conditions is to construct a reasonable discrete monitoring grid system. The present scheme maps the sparsely distributed robot sampling points to regular grid nodes and fills in the data blank area by spatial interpolation, providing a complete and structured data basis for subsequent local gradient estimation.

[0009] By setting grid nodes at key positions such as roadway intersections and turning points, the grid can capture the geometric characteristics of the mine roadway and the key path of disaster propagation; the sparse sampling data is expanded to the entire grid using spatial interpolation techniques, making the originally discrete and discontinuous monitoring data form a continuous field distribution, which is a necessary prerequisite for gradient calculation; the adjacency relationship table clearly shows the spatial topological relationship between nodes, providing a calculation framework for subsequent calculation of the risk difference and directional gradient between adjacent nodes. This conversion from sparse sampling to structured grid essentially regularizes irregular point cloud data, allowing accurate understanding of the spatial variation characteristics of the disaster field under conditions of sparse data through local gradient estimation.

[0010] Further, S3, constructing a disaster risk assessment model at discrete monitoring points, including: for gas disasters, based on the gas concentration and oxygen concentration of the monitoring points, using a piecewise function to calculate the gas risk; for fire disasters, based on the temperature and carbon monoxide concentration of the monitoring points, the fire risk is calculated by weighting; normalize the risk of each grid node for two disasters, map the risk values to the interval of 0 to 1, where 0 means safe and 1 means extremely dangerous; calculate the comprehensive risk of each grid node, adopt the maximum principle or weighted average to fuse the risk values of the two disasters, and generate a comprehensive risk distribution; store the single-disaster risk and comprehensive risk of each grid node as a discrete distribution map, and mark the high-risk nodes whose risk exceeds the threshold. Among them, the risk threshold of high-risk nodes is set to 0.7-0.8, where the gas risk threshold can be 0.7 (corresponding to gas concentration close to the alarm value), and the fire risk threshold can be 0.8 (corresponding to significant abnormalities in temperature or CO concentration).

[0011] In particular, the conversion of disaster parameters with different physical dimensions to dimensionless risk indicators through normalization enables the comparison and operation of gas risk and fire risk on the same numerical scale. This standardization is a necessary condition for effective gradient calculation - only when the risk values of adjacent nodes have the same measurement standard, the risk difference and gradient calculated have physical meaning. Especially under the condition of sparse sampling, the standardized risk values enable accurate reflection of the spatial variation trend of the disaster field through the risk difference between adjacent nodes even with large sampling point spacing, thus achieving effective perception of the disaster evolution trend.

[0012] Further, S4, the spatial variation trend of the hazard degree is calculated by using the adjacent point difference method, including: based on the adjacency relationship table of the grid nodes and the collected wind speed and direction data, a mine ventilation grid topology is constructed, a set of upwind adjacent nodes and a set of downwind adjacent nodes of each node are determined, and the path distance between the current node and each adjacent node along the roadway is calculated; the directed gradient of the gas and fire hazard degree is respectively calculated according to the hazard degree, the path distance and the wind speed weight between the adjacent nodes; the propagation characteristic parameters of the hazard degree of each node are calculated based on the directed gradient, and the propagation characteristic parameters include the upwind propagation intensity and the downwind resistance intensity; the gas hazard degree time variation rate and the fire hazard degree time variation rate of each node are calculated by comparing the hazard degree values at time t and time t-Δt; and the potential coupling risk points are identified according to the propagation characteristic parameters.

[0013] Among them, the adjacent point difference method: in the mine disaster monitoring network, the numerical calculation method for quantitatively describing the spatial distribution gradient of the hazard degree is realized by calculating the difference value of the hazard degree between adjacent grid nodes, and combining the spatial distance and the ventilation direction between the nodes. Adjacent points refer to the grid nodes directly connected to the current node through the roadway; according to the ventilation direction, it is divided into upwind adjacent points and downwind adjacent points; the adjacent relationship is determined by the grid adjacency relationship table.

[0014] In particular, the traditional gradient calculation assumes that the disaster diffuses isotropically in space, ignores the dominant role of the mine ventilation network in disaster propagation, and leads to significant deviation between the prediction results and the actual disaster evolution.

[0015] The application distinguishes the upwind and downwind gradient components by constructing a mine ventilation grid topology, so that the gradient calculation truly reflects the directional propagation characteristics of the disaster under the action of ventilation. Specifically, the wind speed weight is introduced, so that the greater the wind speed of the roadway section, the greater its gradient contribution, accurately depicting the dominant propagation path of disaster materials (such as smoke and gas) along the airflow. By respectively calculating the upwind propagation intensity and the downwind resistance intensity, the system can accurately identify how the fire smoke propagates along the downwind and affects the far-end area, and how the gas accumulates and diffuses against the wind in poor ventilation areas. This directed gradient method not only improves the accuracy of single disaster evolution prediction, but more importantly, it can identify the spatial coupling relationship between the downwind of the fire source and the gas accumulation area.

[0016] Further, the directed gradient of the gas and fire hazard degree is respectively calculated, including: for each grid node i, the hazard degree values of the adjacent nodes are extracted from the upwind node set and the downwind node set respectively; the upwind gradient component is calculated as: wherein R is the hazard degree value, the value range is [0, 1]; is the hazard degree value of the upwind node j; is the tunnel path distance from node i to j, is the wind speed weight, is the measured wind speed between node i and node j, is the maximum wind speed in the ventilation network, is the unit vector from node j to node i, indicating the direction of the upwind gradient;

[0017] Calculate the downwind gradient component: where, is the hazard value of the downwind node k, is the hazard value of the current node, is the tunnel path distance from node i to k, is the wind speed weight of the corresponding path, is the measured wind speed between node i and k, is the unit vector from node i to node k, indicating the direction of the downwind gradient;

[0018] Perform the above gradient calculation for gas hazard and fire hazard respectively to obtain the gas directed gradient and the fire directed gradient ; where the superscript gas represents gas, fire represents fire, up represents upwind, and down represents downwind.

[0019] In particular, in the practical application of mine disaster monitoring, the number of robots is limited and the accessibility of the roadway is restricted, so the robots can usually only cover 5% to 10% of the roadway nodes. This highly sparse sampling poses a serious challenge to disaster field reconstruction. Traditional interpolation methods will produce serious distortion under such sparse sampling conditions, especially they cannot capture the local mutation and directional propagation characteristics of disasters.

[0020] Traditional methods only use the hazard scalar value of the sampling point, while the directed gradient method in this application obtains hazard values , gradient vectors and and the spatial distribution of the gradient, increasing the amount of information provided by each sampling point by 3-5 times, effectively compensating for the sparsity of sampling. More importantly, the differential calculation of upwind and downwind gradient components accurately depicts key physical processes such as the buoyancy effect of gas, the thermal convection of fire, and the transport effect of ventilation, so that even if adjacent sampling points are 100-200 meters apart, the system can still infer the hazard distribution trend in the intermediate region through gradient information.

[0021] Wind speed weight The introduction of the gradient further enhances the adaptability of the method. In high wind speed areas, the gradient mainly reflects the convective transport process, and the danger spreads quickly; while in low wind speed areas, the gradient mainly reflects the diffusion process, and the danger spreads slowly. This adaptive mechanism plays a key role in the main roadway where the sampling is sparse but the wind speed is high. The system can accurately infer the long-distance spread of danger through strong gradients.

[0022] In terms of multi-disaster coupling risk identification, the independent gradient calculation of gas and fire fully considers the different propagation characteristics of the two disasters - gas is easy to accumulate in low-lying places driven by density, while fire shows temperature-driven smoke upward. When the fire upwind blocking strength of a node exceeds the threshold and the gas downwind propagation strength is greater than zero, even if the node has no direct sampling at the moment, the system can identify it as a potential coupling risk point through the gradient feature.

[0023] Further, based on the directed gradient, the propagation characteristic parameters of the danger degree of each node are calculated, including: calculating the downwind propagation strength , which represents the ability of node i to propagate danger in the downwind direction: , , respectively corresponding to the downwind propagation strength of gas and fire; wherein, represents the downwind propagation strength of gas; represents the downwind propagation strength of fire; represents the modulus (Euclidean norm) of the downwind gradient component of gas, and the same below.

[0024] Calculating the upwind blocking strength , which represents the degree of influence of node i by the upwind danger source: , , respectively corresponding to the upwind blocking strength of gas and fire; wherein, represents the upwind blocking strength of gas, represents the upwind blocking strength of fire;

[0025] In particular, the present application converts sparse sampling information into a continuous risk propagation field through the quantitative representation of the downwind propagation strength and the upwind blocking strength .

[0026] The downwind propagation strength quantifies the ability of danger to spread from the current node to the downstream, while the upwind blocking strength which reflects the threat degree of the upstream hazard source to the current node. This bidirectional representation not only captures the static distribution of the hazard, but more importantly, depicts the dynamic propagation process of the hazard. Under the condition of sparse sampling, even if a certain area is not directly monitored by the robot, the system can still infer the risk state of the area through the propagation feature parameters of the adjacent nodes. For example, when a node is detected to have a high value, even if there is no sampling in the downwind direction, the system can predict the propagation path and impact range of the fire smoke.

[0027] The present application simplifies the complex three-dimensional flow field problem into one-dimensional propagation intensity calculation. By calculating the modulus of the gradient component, the system automatically integrates the propagation information in multiple directions, avoiding the difficulty of complete flow field reconstruction under sparse sampling.

[0028] Further, according to the propagation feature parameters, potential coupling risk points are identified, including: judging the fire influence condition: when the fire upwind resistance intensity of node i is greater than a preset threshold , it indicates that the corresponding node is significantly affected by the upwind fire source; wherein, the value range is 0.3 to 0.5; judging the gas accumulation condition: when the gas hazard degree of node i is in the explosion limit range [5%, 16%], and the gas time change rate , it indicates that there is a risk of gas accumulation; when node i satisfies the fire influence condition and the gas accumulation condition at the same time, and the fire time change rate , it is marked as a potential coupling risk point; the coupling risk point is a high-risk location where the interaction of fire and gas may cause secondary disasters (gas explosion). In these locations, fire provides the ignition source, gas provides the explosive material, and the ventilation system becomes the medium for coupling of the two.

[0029] In particular, the calculation of the propagation feature parameter makes full use of the topological characteristics of the mine ventilation network. Under the traditional assumption of uniform sampling, dense monitoring points are needed to discover local hazard coupling; while the present scheme, through the calculation of the directed gradient, makes each sampling point become an "information amplifier", whose propagation feature parameter can reflect the risk state of a larger range upstream and downstream. This spatial correlation of information compensates for the sparsity of sampling, so that a sampling coverage rate of 5% to 10% can achieve a coupling risk identification accuracy rate of more than 90%.

[0030] Further, S5, based on the connectivity analysis of the monitoring network, identifies the propagation path of the hazard degree, including: based on the grid node adjacency relationship table, constructing a directed graph of the monitoring network wherein, the node set V corresponds to the grid nodes, the directed edge set E is determined according to the wind direction, and the edge weight is the propagation strength between nodes; the node with a hazard degree greater than a threshold value is defined as a high-hazard-degree node, the influence range of the high-hazard-degree node is searched by using a Dijkstra algorithm, and the path weight in the Dijkstra algorithm is: wherein, is the downwind propagation strength, is the path segment length; when the cumulative propagation strength from the source node s to the target node t is greater than a threshold value, the corresponding path is marked as an effective propagation path; the cumulative propagation strength is defined as the product of the propagation strength of each node on the path attenuated by: wherein, λ is an attenuation coefficient; a hazard degree propagation tree structure is generated with each high-hazard-degree node as a root, the branch of the tree represents a possible propagation direction, the node depth represents a propagation distance, and the branch weight represents a propagation probability; for the identified potential coupling risk point, reverse search and forward search are performed in the directed graph G: the reverse search finds the fire high-hazard-degree node along the upwind direction edge to determine the fire source propagation path; the forward search finds the node with increased gas concentration along the downwind direction edge to determine the gas diffusion path; and the influence range of the coupling risk is calculated by path superposition analysis.

[0031] Further, S6, according to the hazard degree distribution map and the directed gradient, a risk-avoiding path planning algorithm is used to schedule the robot group, including: according to the hazard degree distribution map and the calculated directed gradient, a dynamic risk map is constructed, the high-hazard-degree node, the potential coupling risk point, and the fire source propagation path and the gas diffusion path are marked as forbidden areas; the dynamic risk map is divided into a safe area, a dangerous area edge, and a high-hazard-area according to the comprehensive hazard degree value; a task priority is set for each robot: the robot in the safe area executes a detection task, the robot in the dangerous area edge executes a monitoring task, and the robot in the high-hazard-area executes an evacuation task; an A* algorithm is used to plan a risk-avoiding path for each robot; a task scheduling instruction is generated and issued to each robot, and the instruction includes a target position and an avoidance strategy.

[0032] Another aspect of the present application also provides a robot group detection scheduling system, comprising: a data acquisition module, which acquires real-time sensing information of a disaster site through a robot group, the real-time sensing information comprising environmental parameters, disaster characteristic data and spatial position data; wherein the environmental parameters comprise wind speed and direction data; a scene reconstruction module, which reconstructs a scene according to the acquired real-time sensing information, generates a three-dimensional digital model of the disaster site, and establishes a discrete monitoring grid based on sampling points; a hazard degree evaluation module, which constructs a disaster hazard degree evaluation model on the discrete monitoring points, calculates the hazard degree values of each monitoring point using measured data for gas and fire, and forms a discrete hazard degree distribution map; a gradient calculation module, which calculates the spatial variation trend of the hazard degree using a neighboring point difference method, and calculates a directed gradient through the hazard degree difference and distance of adjacent monitoring points; when the hazard degrees of any two disasters increase at the same time, the corresponding grid is marked as a potential coupling risk point; a connectivity analysis module, which identifies the propagation path of the hazard degree based on connectivity analysis of the monitoring grid; and a path planning module, which dispatches the robot group using a risk avoidance path planning algorithm according to the hazard degree distribution map and the directed gradient.

[0033] Compared with the prior art, the present application has the following advantages:

[0034] The present application establishes a discrete monitoring grid between sparse sampling points, uses a spatial interpolation technique to supplement data blank areas, calculates a directed gradient of the hazard degree using a neighboring point difference method and wind speed weight, accurately estimates the spatial evolution trend of the disaster, identifies potential coupling risk points by analyzing the spatiotemporal correlation of fire propagation and gas accumulation, predicts the disaster influence range based on propagation characteristic parameters and connectivity analysis, and finally realizes adaptive security deployment of the robot group according to a dynamic risk map.

[0035] The following can be achieved: (1) accurately reconstruct the spatial distribution and evolution trend of the disaster field under sparse sampling conditions through local gradient estimation; (2) effectively identify the formation conditions and development trend of gas-fire coupling risks; (3) predict the disaster propagation path based on limited discrete monitoring data; (4) dynamically optimize the spatial deployment of the robot group to maximize the monitoring coverage range under the premise of ensuring safety; and (5) provide reliable technical support for mine compound disaster situation awareness under sparse data conditions. BRIEF DESCRIPTION OF DRAWINGS

[0036] The present application will be further described in the form of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting, and in these embodiments, the same reference numbers represent the same structures, wherein:

[0037] Figure 1 is an exemplary flowchart of a robot group detection scheduling method according to some embodiments of the present application;

[0038] Figure 2 This is an exemplary flowchart illustrating the construction of a discrete monitoring grid according to some embodiments of this application;

[0039] Figure 3 This is an exemplary flowchart illustrating the creation of a discrete distribution map according to some embodiments of this application;

[0040] Figure 4 This is an exemplary flowchart illustrating the acquisition of coupling risk points according to some embodiments of this application;

[0041] Figure 5 This is an exemplary flowchart illustrating a method for obtaining the scope of influence of coupling risk according to some embodiments of this application. Detailed Implementation

[0042] The methods and systems provided in the embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0043] like Figure 1 As shown, a swarm of robots collects real-time sensing information from the disaster site, including environmental parameters, disaster characteristic data, and spatial location data. Environmental parameters include wind speed and direction data. Based on the collected real-time sensing information, the scene is reconstructed to generate a 3D digital model of the disaster site, and a discrete monitoring grid based on sampling points is established. A disaster hazard assessment model is constructed on these discrete monitoring points. For gas and fire hazards, the hazard values ​​of each monitoring point are calculated using measured data, forming a discrete hazard distribution map. The spatial trend of hazard variation is calculated using the nearest neighbor difference method, and the directed gradient is calculated using the hazard difference and distance between adjacent monitoring points. When the hazard of any two hazards increases simultaneously, they are marked as potential coupled risk points. Based on connectivity analysis of the monitoring network, the propagation path of hazard is identified. Based on the hazard distribution map and the directed gradient, a hazard avoidance path planning algorithm is used to schedule the robot swarm.

[0044] S1 collects real-time perception information from the disaster site through a swarm of robots. The real-time perception information includes environmental parameters, disaster characteristic data, and spatial location data; among which, environmental parameters include wind speed and wind direction data.

[0045] like Figure 2 As shown, S2 reconstructs the scene based on the collected real-time perception information, generates a three-dimensional digital model of the disaster site, and establishes a discrete monitoring grid based on sampling points. This includes matching the robot's own relative positioning data with the pre-stored mine roadway topology to achieve precise determination of absolute coordinates. The robot obtains relative position information through an inertial navigation unit (IMU), odometry, and UWB positioning beacon, and then registers it with the mine's global coordinate system.

[0046] Each spatial sample contains six-tuple information: {coordinate (x, y, z), timestamp t, gas concentration, temperature, wind speed and direction, other sensor data}. The system builds a hash map with spatial coordinates as keys to quickly index the corresponding perception data. A hierarchical attribute storage structure is established, with the base layer storing raw sensor data and the derived layer storing calculated advanced attributes such as hazard level and gradient, supporting fast attribute query and update.

[0047] The system adopts an adaptive grid generation strategy to dynamically adjust the grid density according to the geometric complexity of the roadway: straight roadway section: set a grid node every 10-20 meters along the axial direction of the roadway; intersection and turning point: densify the grid, with a reduced spacing of 5 meters; vertical direction: set 2-3 layers of grid nodes according to the height of the roadway. Each grid node records its type (normal node / key node), the ID of the roadway section it belongs to, and the list of adjacent nodes, forming a complete grid topology.

[0048] For the characteristics of sparse sampling, the system adopts an improved interpolation algorithm based on inverse distance weighting (IDW): adaptively determine the interpolation radius according to the sampling density; interpolation is only performed within connected roadways to avoid cross-roadway interpolation; the weight along the roadway direction is greater than the vertical direction. For grid node i with no sampling data, its attribute value is obtained by interpolation of adjacent sampling points: ; where the weight , is the distance, p is the power index (usually 2), and is the roadway connectivity coefficient.

[0049] Based on the physical connectivity of the roadway and the grid topology, three types of adjacency relationships are defined: direct adjacency: adjacent grid nodes sharing a roadway section; wind flow adjacency: considering the upstream and downstream relationship of wind direction; extended adjacency: second-order adjacency for large-scale gradient calculation. Adjacency table is used for storage, each node maintains: adjacent node ID list; adjacency type label; connected roadway section ID; node distance.

[0050] As shown in Figure 3 S3, a disaster hazard level evaluation model is constructed on discrete monitoring points, and for gas and fire, the hazard level values of each monitoring point are calculated using measured data to form a discrete hazard level distribution map;

[0051] S31, for gas disasters, based on the gas concentration, oxygen concentration, and wind speed data of the monitoring points, a piecewise function is used to calculate the gas hazard level: when the gas concentration is below the safety threshold, the hazard level is 0, and between the safety threshold and the lower explosive limit, it increases linearly, and reaches the maximum value when reaching the explosive concentration range;

[0052] S32, for fire disasters, calculates the fire hazard level through a weighted comprehensive evaluation based on the temperature, carbon monoxide concentration and smoke concentration data of the monitoring points. Among them, the abnormal rise in temperature is given the highest weight, and carbon monoxide and smoke concentration are used as auxiliary indicators.

[0053] S33, for flood disasters, calculates the flood risk based on the water level height, humidity and water pressure data of the monitoring points, combined with the roadway floor elevation information, taking into account the water accumulation rate and possible water inrush paths;

[0054] S34, normalize the three types of hazard risk for each grid node, and map each hazard value to the interval [0, 1], where 0 represents safe and 1 represents extremely dangerous;

[0055] S35, calculate the comprehensive hazard of each grid node, and use the maximum value principle or weighted average method to fuse the hazard values ​​of the three hazards to generate a comprehensive hazard distribution;

[0056] S36 stores the single hazard risk and comprehensive risk of each grid node as a discrete distribution map, and marks high-risk nodes whose risk exceeds a preset threshold, forming a hierarchical risk distribution visualization result.

[0057] like Figure 4 As shown in Figure S4, the spatial variation trend of hazard is calculated using the nearest neighbor difference method. The directed gradient is calculated by the hazard difference and distance between adjacent monitoring points. When the hazard of any two disasters increases simultaneously, they are marked as potential coupled risk points.

[0058] S41. Based on the adjacency table of grid nodes and the wind speed and direction data collected in S1, construct the mine ventilation network topology, determine the upwind adjacent node set and downwind adjacent node set of each node, and calculate the path distance along the roadway between the current node and each adjacent node.

[0059] S42, directional gradient calculations are performed separately for gas and fire hazard levels: for node i, the upwind gradient... Where j is the upwind node, Wind speed weights; downwind gradient , where k is the downwind node;

[0060] S43, Calculate propagation characteristic parameters based on directed gradient: Calculate downwind propagation intensity Characterizes the ability of node i to propagate danger downwind: , These correspond to the downwind propagation intensity of gas and fire, respectively.

[0061] The methane hazard level is calculated from the methane and oxygen concentrations. Specifically... ;

[0062] wherein the gas concentration Segmented function:

[0063]

[0064] Oxygen concentration correction coefficient :

[0065]

[0066] represents the fire hazard value, which is calculated by temperature and carbon monoxide concentration, ; wherein, , is a weight coefficient.

[0067] Temperature hazard function :

[0068]

[0069] Carbon monoxide hazard function :

[0070] .

[0071] Calculate the upwind barrier strength , which represents the degree of influence of the upwind hazard source on node i: , , respectively corresponding to the upwind barrier strength of gas and fire;

[0072] Calculate the comprehensive propagation factor: wherein, is a weight coefficient, reflecting the importance of upwind threat, with a value range of 0.3 to 0.5;

[0073] According to the spatio-temporal distribution of propagation characteristic parameters, identify hazard source nodes (nodes with value is large), threatened nodes (nodes with value is large) and propagation channel nodes (nodes with and are both large);

[0074] Establish the mapping relationship between propagation characteristic parameters and grid nodes, providing quantitative basis for subsequent risk assessment.

[0075] S44, by comparing the hazard value at time t and time t-Δt, calculate the gas hazard time change rate and the fire hazard time change rate of each node; Δt represents the sampling time interval, typical value 30-60 seconds;

[0076] S45, Identifying potential coupling risk points based on propagation characteristic parameters: Determining the fire source influence conditions: When the fire backwind barrier strength at node i... Greater than the preset threshold This indicates that the corresponding node is significantly affected by the upwind fire source;

[0077] Determine the condition for gas accumulation: when the gas hazard level of node i... Within the explosive limits [5%, 16%], and the rate of change of gas over time... This indicates a risk of gas accumulation;

[0078] Determine the coupling risk condition: When node i simultaneously satisfies the fire source influence condition and the gas accumulation condition, and the fire time change rate... When this occurs, it is marked as a potential coupling risk point;

[0079] Calculate the coupling risk strength: ,in, This reflects the potential chain reaction caused by the downstream spread of gas;

[0080] Based on the strength of coupling risk Classify potential coupling risk points. (0.7) indicates high risk. Medium risk (0.3) represents low risk, and a hierarchical coupling risk distribution map is generated.

[0081] S46, Calculate the coupling risk level: For the marked coupling risk points, calculate their risk level. Among them, the basic risk weight of gas Fire basic risk weights Coupling effect weights Propagation and diffusion weights This generates a hazard distribution map that includes gradient information and coupled risk levels.

[0082] like Figure 5 As shown in Figure S5, based on connectivity analysis of the monitoring network, the propagation path of the hazard is identified; the discrete monitoring grid is converted into a directed graph. This achieves the mapping from physical space to a graph theory model; the node set V directly corresponds to the grid nodes, inheriting all monitoring data and hazard information; the direction of the directed edge set E is determined by the wind direction: if the wind blows from node i to node j, then a directed edge (i, j) is established; edge weights... (node i's downwind propagation strength), reflecting the ability of the disaster to propagate along the edge; the directed graph model accurately depicts the one-way propagation characteristics of the mine ventilation network, avoiding the unreasonable assumption of disaster propagation against the wind in the undirected graph model. The edge weight uses the propagation strength rather than the distance, allowing the graph search algorithm to find the "easiest path" rather than the "shortest path" of disaster propagation.

[0083] Nodes with a hazard greater than the threshold value (0.7) are defined as high-risk nodes, and Dijkstra's algorithm is used to search the influence range of high-risk nodes. The path weight in Dijkstra's algorithm is: wherein, is the downwind propagation strength, is the path segment length; the traditional Dijkstra finds the shortest path, and the weight inverse design of the present application makes the path weight smaller when the propagation strength is greater, making it easier to find the "smallest resistance" propagation path. When , , the low propagation strength path is automatically blocked, avoiding non-physical long-distance propagation.

[0084] When the cumulative propagation strength from the source node s to the target node t is greater than the threshold value, the corresponding path is marked as an effective propagation path; the cumulative propagation strength is defined as the product of the propagation strength of each node on the path, and the attenuation is: wherein λ is the attenuation coefficient; the product form is used to reflect the cascade attenuation characteristics of propagation.

[0085] Perform a breadth-first search with each high-risk node as the root; only expand child nodes along effective propagation paths; record the depth (hop count) and cumulative strength to the root of each node; the weight of each branch is set to the propagation probability of the path, and the calculation formula is: wherein the denominator is the sum of the cumulative strengths of all branches; use the parent-child relationship table to store the tree structure; node attributes include: depth, cumulative strength, and propagation probability; support fast subtree traversal and path backtracking.

[0086] For the identified potential coupling risk points, perform a reverse search and a forward search in the directed graph G:

[0087] Starting from the coupling risk point, perform a reverse search along the upwind edge; find nodes that satisfy as potential fire sources; search depth limit: 5 hops or cumulative distance 500m; record all possible fire-coupling point propagation paths.

[0088] Starting from the coupling risk point, perform a forward search along the downwind edge; find and nodes; these nodes may become new coupling risk points; form a "chain of infection" of coupling risks.

[0089] Superimpose the fire source propagation path and the gas diffusion path in space; the intersection area is the coupling risk influence range; calculate the comprehensive risk value of each node: .

[0090] S6, according to the risk degree distribution map and the directed gradient, the risk avoidance path planning algorithm is used to schedule the robot group. According to the generated risk degree distribution map and the calculated directed gradient, a dynamic risk map is constructed, and the high-risk nodes marked, the potential coupling risk points identified, and the fire source propagation path and the gas diffusion path determined are marked as forbidden areas;

[0091] Based on the comprehensive risk degree value, the region is divided: the node region with a comprehensive risk degree less than 0.3 is defined as a safe region, the node region with a comprehensive risk degree between 0.3 and 0.7 is defined as a dangerous region edge, and the node region with a comprehensive risk degree greater than 0.7 is defined as a high-risk region;

[0092] Set the task priority for each robot: the robot in the safe region performs a detection task, the robot in the dangerous region edge performs a monitoring task, and the robot in the high-risk region performs an evacuation task;

[0093] An A* algorithm is used to plan a risk avoidance path for each robot, and the comprehensive risk degree and propagation characteristic parameters of the node are added as risk weights in the path cost function, and a path away from the forbidden area is preferentially selected;

[0094] Generate task scheduling instructions and issue them to each robot, and the scheduling instructions include: robot ID, task type, priority; target position coordinate sequence (way points); predicted arrival time sequence; avoidance strategy parameters (waiting points, yielding objects); emergency plan (alternative path).

[0095] The above describes the application creation and its implementation mode in a schematic manner, which is not restrictive, and the application can be realized in other specific forms without departing from the spirit or essential characteristics of the application. The embodiments shown in the drawings are only one of the embodiments of the application, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired by it, without departing from the spirit of the application, similar structural forms and embodiments can be designed without creative design, which shall belong to the protection scope of the application. In addition, the word "comprising" does not exclude other elements or steps, and the word "one" before an element does not exclude including "multiple" elements. The words "first", "second", etc. are used to represent names, and do not represent any specific order.

Claims

1. A robot swarm detection and scheduling method, characterized in that, include: The robot swarm collects real-time sensing information from the disaster site, including environmental parameters, disaster characteristic data, and spatial location data; among which, environmental parameters include wind speed and direction data. Based on the collected real-time sensing information, the scene is reconstructed to generate a three-dimensional digital model of the disaster site, and a discrete monitoring grid based on sampling points is established. A hazard risk assessment model is constructed at discrete monitoring points. For gas and fire, the hazard value of each monitoring point is calculated using measured data to form a discrete hazard distribution map. The spatial variation trend of hazard is calculated using the nearest neighbor difference method, and the directed gradient is calculated by the hazard difference and distance between adjacent monitoring points; when the hazard of any two hazards increases simultaneously, they are marked as potential coupled risk points. Based on connectivity analysis of the monitoring network, the propagation path of the hazard is identified; Based on the hazard distribution map, directed gradient, and propagation path, a hazard avoidance path planning algorithm is used to schedule the robot swarm.

2. The robot swarm detection and scheduling method according to claim 1, characterized in that: S2, based on the collected real-time sensing information, reconstructs the scene and generates a three-dimensional digital model of the disaster site, including: Based on the spatial location data of the robot swarm and the pre-stored mine roadway topology, the absolute coordinates of each robot in three-dimensional space are determined. The environmental parameters and disaster characteristic data collected by each robot are mapped to corresponding spatial coordinate points to form a set of spatial sampling points with attribute information. Based on the geometric characteristics of the mine roadway, a three-dimensional mesh framework is established along the roadway direction and vertical direction, and the mesh nodes are set at roadway intersections, turning points and equally spaced points. The actual sampling points are mapped to grid nodes. For grid nodes without sampling data, spatial interpolation is performed based on the data of adjacent nodes to obtain discrete monitoring grids. Establish a time-series data cache for each grid node for temporal evolution analysis of hazard level; An adjacency table between grid nodes is established based on the connectivity of the roadways, which is used for gradient calculation and connectivity analysis.

3. The robot swarm detection and scheduling method according to claim 1, characterized in that: S3, Constructing a disaster risk assessment model at discrete monitoring points, including: For gas disasters, a piecewise function is used to calculate the gas hazard level based on the gas and oxygen concentrations at monitoring points. For fire hazards, the fire hazard level is calculated by weighting the temperature and carbon monoxide concentration at the monitoring points. The two types of hazard risk levels for each grid node are normalized, and each risk level value is mapped to the interval between 0 and 1, where 0 represents safe and 1 represents extremely dangerous. Calculate the overall hazard of each grid node, and combine the hazard values ​​of the two hazards using the maximum value principle or weighted average to generate an overall hazard distribution; The individual hazard risk and overall hazard risk of each grid node are stored as a discrete distribution map, and high-risk nodes with hazard levels exceeding the threshold are marked.

4. The robot swarm detection and scheduling method according to claim 2, characterized in that: S4, using the nearest neighbor difference method to calculate the spatial variation trend of hazard, including: Based on the adjacency table of grid nodes and the collected wind speed and direction data, a mine ventilation grid topology is constructed, the upwind adjacent node set and downwind adjacent node set of each node are determined, and the path distance along the roadway between the current node and each adjacent node is calculated. Based on the hazard level, path distance, and wind speed weights between adjacent nodes, the directed gradients of gas and fire hazard levels are calculated respectively. The propagation characteristic parameters of each node’s hazard level are calculated based on the directed gradient. The propagation characteristic parameters include downwind propagation intensity and upwind blocking intensity. By comparing time t and The hazard value at any given time is used to calculate the time-varying rate of gas hazard and fire hazard for each node; Potential coupling risk points are identified based on propagation characteristic parameters and the rate of change of risk over time.

5. The robot swarm detection and scheduling method according to claim 4, characterized in that: Calculate the directed gradients of gas and fire hazard levels separately, including: For each grid node i, start from the upwind node set. and downwind node set Extract the danger values ​​of adjacent nodes; Calculate the upwind gradient component: Where R is the hazard value. Let be the path distance from node i to j in the alleyway. Assuming wind speed as the weighting, To measure the wind speed, This represents the maximum wind speed in the ventilation network. It is a pointer to a unit vector; Calculate the downwind gradient component: ,in, Let k be the hazard value of the downwind node. This represents the danger level of the current node. Let be the path distance from node i to k through the alleyway. The wind speed weights for the corresponding paths, The measured wind speeds between nodes i and k Let be the unit vector pointing from node i to node k; Perform the above gradient calculations on the gas hazard level and fire hazard level respectively to obtain the directional gas gradient. Fire directional gradient The superscript "gas" indicates natural gas, "fire" indicates fire, "up" indicates upwind, and "down" indicates downwind.

6. The robot swarm detection and scheduling method according to claim 4, characterized in that: The propagation characteristic parameters of the danger level of each node are calculated based on the directed gradient, including: Calculate the downwind propagation intensity Characterizes the ability of node i to propagate danger downwind: , These correspond to the downwind propagation intensity of gas and fire, respectively. Calculate headwind blocking strength This characterizes the degree to which node i is affected by the upwind hazard source: , These correspond to the backwind resistance strength for gas and fire, respectively.

7. The robot swarm detection and scheduling method according to claim 6, characterized in that: Potential coupling risk points are identified based on propagation characteristic parameters, including: Determine the conditions for determining the impact of the fire source: When the fire backwind barrier strength at node i Greater than the preset threshold This indicates that the corresponding node is significantly affected by the upwind fire source; Determine the condition for gas accumulation: when the gas hazard level of node i... Within the explosive limits [5%, 16%], and the rate of change of gas over time. This indicates a risk of gas accumulation; When node i simultaneously satisfies both the fire source influence condition and the gas accumulation condition, and the fire time variation rate When this occurs, it is marked as a potential coupling risk point.

8. The robot swarm detection and scheduling method according to any one of claims 2 to 7, characterized in that: S5, based on connectivity analysis of the monitoring network, identifies the propagation paths of hazard, including: A directed graph of the monitoring network is constructed based on the adjacency table of grid nodes. In this context, the node set V corresponds to the grid nodes, the directed edge set E is determined according to the wind direction, and the edge weight is the propagation intensity between nodes. Nodes with a risk level greater than a threshold are defined as high-risk nodes. Dijkstra's algorithm is used to search for the influence range of high-risk nodes. Path weights in Dijkstra's algorithm are as follows: ,in, For the intensity of downwind propagation, This represents the length of the path segment. Depending on the scope of impact, when from the source node To the target node When the cumulative propagation intensity exceeds a threshold, the corresponding path is marked as a valid propagation path; the cumulative propagation intensity is defined as the product decay of the propagation intensities of each node on the valid propagation path: , where λ is the attenuation coefficient; For the identified potential coupling risk points, perform reverse search and forward search in the directed graph G: Based on the effective propagation path, a reverse search is conducted along the upwind side to find high-risk fire nodes and determine the fire source propagation path. Based on the effective propagation path, a forward search is conducted along the downwind side to find nodes where the gas concentration increases, thus determining the gas diffusion path. The impact range of coupling risk is calculated through path overlay analysis.

9. The robot swarm detection and scheduling method according to claim 8, characterized in that: S6. Based on the hazard distribution map and directed gradient, a hazard avoidance path planning algorithm is used to schedule the robot swarm, including: Based on the hazard distribution map and the calculated directed gradient, a dynamic risk map is constructed, and high-hazard nodes, potential coupling risk points, fire source propagation paths, and gas diffusion paths are marked as prohibited areas. Based on the comprehensive risk level, the dynamic risk map is divided into safe zones, dangerous zone edges, and high-risk zones. Assign task priorities to each robot: robots in safe areas perform detection tasks, robots at the edge of dangerous areas perform monitoring tasks, and robots in high-risk areas perform evacuation tasks. The A* algorithm is used to plan obstacle avoidance paths for each robot; Generate task scheduling instructions and send them to each robot. The instructions include the target location and avoidance strategy.

10. A robot swarm detection and scheduling system, characterized in that, include: The data acquisition module collects real-time sensing information from the disaster site through a swarm of robots. This real-time sensing information includes environmental parameters, disaster characteristic data, and spatial location data; among which, environmental parameters include wind speed and direction data. The scene reconstruction module reconstructs the scene based on the collected real-time sensing information, generates a three-dimensional digital model of the disaster site, and establishes a discrete monitoring grid based on sampling points. The hazard assessment module constructs a hazard hazard assessment model at discrete monitoring points. For gas and fire, it uses measured data to calculate the hazard value of each monitoring point and forms a discrete hazard distribution map. The gradient calculation module uses the nearest neighbor difference method to calculate the spatial variation trend of hazard level, and calculates the directed gradient by the hazard difference and distance between adjacent monitoring points; when the hazard levels of any two disasters increase simultaneously, the corresponding grid is marked as a potential coupled risk point. The connectivity analysis module identifies the propagation paths of hazards based on the connectivity analysis of the monitoring grid. The path planning module schedules the robot swarm using a risk avoidance path planning algorithm based on the risk distribution map and directed gradient.

Images