A mine safety fog computing edge node lightweight deployment method and system

Abstract

The application relates to the technical field of edge computing, in particular to a mine safety fog computing edge node lightweight deployment method and system. The method comprises the following steps: constructing a mine three-dimensional grid model and marking existing sensors and fog nodes, calculating the edge node layout fitness of each grid by analyzing the local and global sensor distribution density and communication distance characteristics; secondly, the grid is analyzed by using the fitness and position difference, and the actual deployment grid set is preliminarily determined by combining the inter-cluster coverage similarity screening to eliminate redundancy; finally, an optimization model is constructed in the candidate point set, the optimization model has the double targets of minimizing the total deployment number and minimizing the load imbalance degree, and the final optimal deployment position of the edge node is obtained through an optimization algorithm. The application aims to realize the lightweight deployment of the mine edge node.

Description

Technical Field

[0001] This application relates to the field of edge computing technology, specifically to a lightweight deployment method and system for fog computing edge nodes in mine safety. Background Technology

[0002] Mine production data is a key basis for assessing its safety status and exploitable potential. For example, the predicted amount of carbon monoxide concentration underground is directly related to the safety of miners, and the predicted amount of minerals will affect subsequent mining plans. Fog computing is a distributed computing architecture, usually deployed between edge nodes and cloud centers. Due to the wide geographical distribution of mines, the data transmission distance between the monitoring end and the cloud is large, resulting in a large latency in the data obtained by the cloud server.

[0003] To address this issue, lightweight deployment of edge nodes for mine fog computing is crucial. However, existing deployment methods often employ optimization models, directly deploying multiple nodes collaboratively based on pre-defined computing power requirements and other constraints. This approach has significant shortcomings: Firstly, it lacks refined quantitative analysis of sensor density and fog node communication radius, leading to blind site selection, monitoring blind spots, or excessive redundant coverage, resulting in wasted hardware resources and increased costs. Secondly, it neglects the dynamic balancing of data load among nodes, easily causing some edge nodes to become overloaded and become communication bottlenecks, leading to delays or even loss of secure data transmission, failing to meet the urgent needs of mine disaster early warning for real-time performance, reliability, and lightweight deployment. Summary of the Invention

[0004] In view of the above, it is necessary to provide a lightweight deployment method and system for edge nodes of fog computing in mine safety to solve the above problems.

[0005] The first aspect of this application provides a lightweight deployment method for edge nodes in mine safety fog computing, the method comprising: Collect 3D point cloud data of the mine to build a mine mesh model and mark the sensors and fog nodes already deployed in the mine; Analyze the sensor distribution density of each grid relative to its neighboring grids and the global grid to obtain the sensor distribution characteristics of each grid; within the preset maximum communication radius of the edge node device, analyze the distance characteristics between each grid and the grid where the sensor is located and the grid where the fog node is located, and combine the sensor distribution characteristics to calculate the edge node deployment adaptability of each grid. Based on the differences in edge node deployment fitness and location among the grids, all grids are clustered. Based on the overall distribution characteristics of edge node deployment fitness and the number of grids in each cluster, the grids with deployed edge nodes are initially labeled. The sensor coverage similarity of labeled grids among different clusters is analyzed to determine the actual set of grids with deployed edge nodes. Within the grid set of actual deployed edge nodes, candidate edge nodes are preset. With the optimization objectives of minimizing the total number of edge node deployments and minimizing the load imbalance between nodes, a multi-objective optimization model is constructed to optimize the deployment of candidate edge nodes, resulting in the final set of edge node deployment locations.

[0006] Preferably, the acquisition of sensor distribution characteristics for each grid specifically involves: Obtain the normalized result of the number of sensors in each grid under all its neighboring grids, and obtain the ratio of the number of sensors in each grid and its neighboring grids to the total number of sensors; multiply the obtained normalized result with the ratio to obtain the first local feature; average the obtained first local feature with the normalized result of the number of sensors in each grid under all grids to obtain the sensor distribution feature of each grid.

[0007] Preferably, the calculation of the edge node placement fitness of each grid is specifically as follows: For the i-th grid L(i), with the geometric center of L(i) as the reference, within the maximum communication radius of the edge node device, the maximum connected path distance between L(i) and all the grids where the sensors are located within this range is used as the data source access loss radius of L(i); the minimum connected path distance between L(i) and the grid where the nearest deployed fog node is located within this range is used as the uplink backhaul distance of L(i). Obtain the negative correlation mapping result of the sum of the uplink backhaul distance and the data source access loss radius of L(i), and perform positive fusion with the uplink backhaul distance and the sensor distribution characteristics to obtain the edge node deployment fitness of L(i).

[0008] Preferably, the distance metric used to cluster all grids is: The normalized geometric center coordinates of each grid and the fitness of edge node placement are concatenated to construct a four-dimensional feature vector, and the distance of the connected paths between the four-dimensional feature vectors is used as the clustering metric.

[0009] Preferably, the initial marking of the mesh for deploying edge nodes specifically involves: Clusters are filtered based on the overall distribution of edge node deployment fitness across all grids. Within each filtered cluster, a representative grid is established based on the maximum edge node deployment fitness. Simultaneously, intra-cluster coverage deduplication is performed using the effective communication radius of the edge node devices. Finally, a grid set with preliminarily labeled edge node deployments is output. The effective communication radius is obtained by multiplying the maximum communication radius of the edge node devices by a preset reduction factor.

[0010] Preferably, determining the mesh set for the actual deployment of edge nodes specifically involves: Calculate the similarity of sensors that can communicate with each other within the grid that are paired and marked as deployment edge nodes; use the mean of the sensor similarity of all pairs of grids marked as deployment edge nodes as the first judgment threshold. For two grids whose spatial distance is less than the maximum communication radius of the edge node device and whose sensor similarity is greater than the first judgment threshold, retain the grid with the highest edge node deployment fitness and remove the other grid. After traversing all combinations, the set of all the retained grids is taken as the grid set for actual deployment of edge nodes.

[0011] Preferably, the objective function of the multi-objective optimization model is: Each candidate edge node in the actual deployed edge node grid is assigned a decision variable. When the first When a candidate point is actually selected for deployment of an edge node ;otherwise ; Minimize the total number of deployments: ;in, This refers to the set of all candidate edge nodes generated within all meshes marked as deployable edge nodes. Used to achieve lightweight deployment; Minimize load imbalance: ;in, To prevent extremely small positive numbers with a denominator of zero; For global average load, The actual load borne by the selected nth edge node candidate point; Used to minimize load imbalance between edge nodes.

[0012] Preferably, a load-sharing strategy within the same grid is adopted, i.e., if candidate points Belongs to the grid and ,but , For grid The set of all candidate edge nodes within the region. This represents the total load that needs to be processed in the g-th grid. Let represent the decision variable for the k-th edge node candidate point.

[0013] Preferably, the constraints for all multi-objective optimization models are as follows: First constraint: ;in, For the actual mesh set where edge nodes are deployed, For set Grid in ; For grid The set of all candidate edge nodes within the region; The second constraint: ;in, The maximum load supported by a single edge node hardware device.

[0014] Secondly, embodiments of this application also provide a lightweight deployment system for fog computing edge nodes in mine safety, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of any of the methods described above.

[0015] This application has at least the following beneficial effects: This application achieves accurate calculation of edge node deployment fitness by constructing a three-dimensional mesh model of the mine and quantitatively analyzing the distribution density and distance characteristics of sensors, effectively solving the problem of blind node deployment in complex mining environments. By using a clustering algorithm based on the differences in edge node deployment fitness and location characteristics, high-value deployment areas can be identified, and redundant nodes can be eliminated by analyzing the similarity of sensor coverage between clusters. This significantly reduces the initial number of edge node candidates while ensuring full monitoring coverage. This data-driven preliminary screening mechanism not only reduces the complexity of subsequent optimization calculations but also ensures that the deployment location closely matches the actual sensing needs of the mine, laying a solid spatial foundation for lightweight deployment.

[0016] Building upon this foundation, the multi-objective optimization model of the proposed solution focuses on minimizing the total number of deployments and minimizing load imbalance, further achieving globally optimal resource allocation. Through refined optimization within the candidate point set, this strategy can meet the real-time requirements of mine safety monitoring while minimizing hardware investment costs and operational energy consumption, avoiding communication bottlenecks or data loss risks caused by node overload. The final deployment scheme balances economy and efficiency, while also improving the robustness and response speed of the fog computing architecture in harsh mining environments, providing reliable technical support for real-time perception and intelligent early warning of mine safety status. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of a lightweight deployment method for fog computing edge nodes in mine safety, provided as an embodiment of this application. Detailed Implementation

[0018] The following description, in conjunction with the accompanying drawings, details the specific scheme of a lightweight deployment method and system for edge nodes in mine safety fog computing provided in this application.

[0019] Please see Figure 1The diagram illustrates a flowchart of a lightweight deployment method for fog computing edge nodes in mine safety according to an embodiment of this application. The method includes the following steps: The first step is to collect 3D point cloud data of the mine to build a mine mesh model and mark the sensors and fog nodes already deployed in the mine.

[0020] Using an underground inspection robot equipped with lidar, three-dimensional point cloud data of mine roadways and working faces are collected.

[0021] A voxel filtering algorithm is used to filter the acquired 3D point cloud data to eliminate potential noise interference. The noise-free point cloud is then used as input, and MeshLab software is used to reconstruct the input point cloud in 3D. The Poisson reconstruction algorithm is called in the software to mesh the 3D model, resulting in a mine mesh model. The distance between meshes is defined as the shortest distance along the topological path of the mesh geometric center along the mine cavity mesh. At the same time, the sensors in the mine are marked with red squares, and the actual fog nodes are marked with blue triangles.

[0022] Point cloud filtering algorithms and the use of 3D reconstruction software to process point cloud data are well-known technologies in this field, and the specific calculation methods will not be elaborated here.

[0023] The second step is to analyze the sensor distribution density of each grid relative to its neighboring grids and the global grid, and obtain the sensor distribution characteristics of each grid; within the preset maximum communication radius of the edge node device, analyze the distance characteristics between each grid and the grid where the sensor is located and the grid where the fog node is located, and calculate the edge node deployment adaptability of each grid in combination with the sensor distribution characteristics.

[0024] In mining scenarios, the sensor requirements differ across mining and ore transportation processes, leading to variations in the types and quantities of sensors selected. For instance, in mining areas, multiple monitoring sensors, such as carbon monoxide and pressure sensors, need to be densely deployed to ensure safety. However, at mine entrances, due to their complex distribution, the installation positions of cameras need to be adjusted according to the specific environment, resulting in a degree of randomness in their distribution. This differentiated sensor arrangement leads to uneven sensor distribution within and between grids, as well as variations in distribution between adjacent grids and between different grids.

[0025] The deployment of edge nodes needs to take into account the communication distance between them and fog nodes and sensor terminals. The farther away they are from fog nodes and sensor terminals, the more interference they may experience during data transmission, and the greater the data loss will be.

[0026] Based on the above analysis, taking the i-th grid L(i) as an example, its neighborhood is divided. In this embodiment, the grids within a 3×3×3 range centered on L(i) are considered as the neighborhood grids of L(i). This application does not impose special restrictions on the neighborhood range of each grid, which can be adjusted according to the actual situation. In this embodiment, the distance between grids is the distance of the connected path between the centroids of the grids.

[0027] First, the edge node deployment fitness of L(i) is constructed, which is an evaluation index that comprehensively measures the sensor distribution density within the grid, data source access loss, and uplink backhaul efficiency. The specific formula is as follows: In the formula, Indicates the edge node deployment fitness of L(i); The sensor distribution characteristics of L(i) are represented; Indicates the data source access loss radius of L(i); Represents the uplink backpropagation distance of L(i); To represent extremely small positive numbers, and to prevent the denominator from being zero, this embodiment... ; where, through the formula This represents the negative correlation between the on-chain backhaul distance of L(i) and the cumulative sum of the data source access loss radius; expressed by the formula... This represents the negative correlation mapping result between the sum of the on-chain backhaul distance of L(i) and the data source access loss radius, and the result after positive fusion with the on-chain backhaul distance and the sensor distribution characteristics.

[0028] The sensor distribution characteristics of L(i) are calculated as follows: First, the normalized result of the number of sensors in L(i) under all neighborhood grids of L(i) is calculated, and the ratio of the number of sensors in L(i) and its neighborhood grids to the total number of sensors is obtained; the normalized result is multiplied by the ratio to obtain the first local feature; further, the first local feature is averaged with the normalized result (min-max normalization) of the number of sensors in L(i) under all grids to obtain the sensor distribution characteristics of L(i). In the calculation process, if the denominator is zero, a very small positive number is added to the denominator for compensation. In this embodiment, the very small positive number is 0.01.

[0029] The data source access loss radius of L(i) reflects the transmission cost of sensor data entering the edge node. It is obtained as follows: using L(i) as a reference, identify all sensor grids located at distances less than or equal to the maximum communication radius of the edge node device from L(i). Calculate the connection path distance between these grids and L(i), and use the maximum connection path distance as the data source access loss radius of L(i). It should be noted that the edge node devices within each grid are assumed to have the same hardware configuration and communication capabilities, and the maximum communication radius of the edge node devices is assumed to be 50 meters. Calculating the data source access loss radius of each grid is to quantify the transmission delay and signal attenuation generated when the raw sensor data is transmitted to the edge node. A smaller distance indicates a shorter data transmission link, faster response speed at the data source, and lower environmental ionization and interference losses.

[0030] Here, the uplink backhaul distance of L(i) is used to measure the communication cost of uploading the processed data from the edge node to the fog node. It is obtained by: using the geometric center of L(i) as a reference, finding the nearest deployed fog node within the maximum communication radius of the edge node device, and calculating the minimum connected path distance between these two grids, denoted as L(i). The "nearest fog node" refers to the deployed on-chain node, not the edge node to be deployed. Calculating the on-chain backhaul distance is to quantify the bandwidth consumption, signal latency, and interference loss incurred when edge nodes transmit data back to fog nodes. The smaller the on-chain backhaul distance, the shorter the transmission distance between the edge node and the fog node, the higher the data on-chain backhaul efficiency, which can effectively reduce the overall network response latency and improve the transmission stability of the edge computing architecture.

[0031] In the formula for the fitness of edge node deployment, it should be understood that the greater the sensor distribution characteristics of each grid, the better. The larger the value, the denser the sensors within the grid and its neighborhood. Deploying edge nodes here allows for the aggregation of more data sources at once, improving the utilization of computing resources; the denominator... This is the total transmission path length from the sensor to the edge node and then to the fog node (approximately in the case of a linear arrangement). The smaller this term, the lower the overall network latency; numerator The larger the value, the farther the grid location is from the deployed fog nodes. Deploying edge nodes at this location allows for more effective interception of raw data, relieving the remote communication pressure on the fog nodes and demonstrating the value of edge computing deployment. It enables earlier interception and processing of raw data, significantly reducing the amount of invalid data transmitted back to the fog nodes (a core advantage of edge computing: data preprocessing and compression); edge node deployment adaptability. The larger the value, the more likely the grid not only covers dense sensor data sources but also is closer to the data generation end in the network topology while maintaining effective connectivity with upper-layer fog nodes, making it an ideal location for deploying edge computing nodes.

[0032] The third step is to cluster all grids based on the differences in edge node deployment fitness and location among the grids. Based on the overall distribution characteristics of edge node deployment fitness of all grids in each cluster and the number of grids, the grids where edge nodes are deployed are initially marked. The similarity of sensor coverage of marked grids among different clusters is analyzed to determine the actual set of grids where edge nodes are deployed.

[0033] Furthermore, since some sensors are deployed in relatively close areas, two grids with high edge node deployment adaptability are often adjacent or close to each other. When these two grids communicate with each other, the sensors used have a high degree of similarity, so only one grid needs to be retained, thus avoiding data redundancy.

[0034] In this embodiment, the min-max algorithm is used to construct a four-dimensional feature vector by concatenating the normalized geometric center coordinates (front) and edge node fitness (back) of each grid. The distance of the connected paths between the four-dimensional feature vectors is then calculated as the clustering metric. This embodiment uses the DBSCAN clustering algorithm to cluster all grids. The neighborhood radius of the DBSCAN algorithm is set to 0.15, and the minimum containment point value is set to 1, resulting in several clusters.

[0035] First, the arithmetic mean of the fitness of all grid edge nodes in the global range is calculated as the judgment threshold. If the mean fitness of all grid edge nodes in a cluster is less than the judgment threshold, the cluster is marked as a non-important cluster. For clusters with fitness greater than or equal to the judgment threshold, if the number of grids in the cluster is equal to 1, the grids in the cluster are directly marked as grids with deployable edge nodes. Otherwise, the grid with the highest fitness of edge nodes is selected as the representative grid. It should be noted that if there are multiple equal maximum values, the mean of the three-dimensional coordinates of the geometric centers of all grids corresponding to the maximum value is taken as the population centroid, and the grid closest to the population centroid is selected as the representative grid.

[0036] After obtaining the representative grid in each cluster, the connectivity path distance between the remaining grids in the cluster and the representative grid is calculated. If the connectivity path distance between the grid and the representative grid is less than or equal to the effective communication distance of the edge node device (obtained by multiplying the maximum communication distance by a reduction factor to prevent weak edge communication signals from affecting data transmission; this application does not impose special restrictions on the value of the reduction factor, which can be adjusted according to actual conditions; in this embodiment, the reduction factor is 0.9), it indicates that these grids are within the same communication range as the representative grid, and only the representative grid is marked as the grid for deploying edge nodes. For the remaining grids in the cluster (i.e., grids not within the same communication range as the representative grid), the above filtering is repeated for the grid with the highest edge node deployment fitness, and the filtered grids are marked as the grids for deploying edge nodes.

[0037] For different clusters, firstly, the similarity of sensors capable of data communication within each pair of grids marked as deployment edge nodes is calculated: For each grid marked as a deployment edge node, the one-hot encoded vector of the set of all sensors within its maximum communication radius is determined. The average of the ratios of the intersection (i.e., the number of sensors simultaneously covered by both grids) and the union (the total number of sensors covered by both grids after deduplication) of the one-hot encoded vectors of all pairwise combined marked grids is used as the first judgment threshold. During the calculation process, if the denominator is zero, a very small positive number is added to the denominator for compensation. In this embodiment, the value of the very small positive number is 0.01.

[0038] For all grids marked as edge node deployments, calculate the connectivity path distance and sensor similarity between any two grids. If the distance between two grids is less than or equal to the effective communication distance of the edge node device, and the sensor similarity is greater than or equal to a first judgment threshold, compare the edge node deployment fitness, retain the grids with higher edge node deployment fitness, and discard the grids with lower fitness. After traversing all combinations, the set of all retained grids is the final set of grids for actual edge node deployment.

[0039] The fourth step is to preset candidate edge nodes within the grid set of actual deployed edge nodes, and construct a multi-objective optimization model to optimize the deployment of candidate edge nodes with the optimization objectives of minimizing the total number of edge node deployments and minimizing the load imbalance between nodes, so as to obtain the final set of edge node deployment locations.

[0040] First, based on the deployment characteristics of historical edge nodes, 3D reconstruction software is used to deploy candidate edge nodes in all grids in the mine that can be used as actual deployed edge nodes, and these candidate points are marked with green circles. In this embodiment, the geometric center of the grid, the centroid of the sensor cluster within the grid, and the apex corner of the grid are all used as candidate edge nodes.

[0041] Secondly, we construct the first objective function, which aims to minimize the total number of candidate edge node deployments to achieve lightweight deployment of edge nodes. Its formula is: in, This represents the total number of all actually deployed edge nodes; Let p be the decision variable for the p-th candidate edge node. The range is 1 or 0. A value of 1 indicates that the candidate edge node is selected, meaning that the edge node hardware device needs to be installed at this location. A value of 0 indicates that the candidate edge node was not selected and does not require installation; p represents the p-th candidate edge node. This represents the set of all candidate edge nodes generated within all grids marked as deployable edge nodes.

[0042] Edge nodes all have an upper limit to their hardware processing capabilities. By minimizing the deviation between the load of each edge node and the average load, it is possible to effectively prevent some nodes from overheating prematurely, losing data, or even crashing, thereby improving the robustness of the entire mine safety monitoring system. When the load of each edge node is balanced, it has better resilience and scalability when facing sudden surges in massive amounts of mine sensor data. Based on this, a second objective function is constructed, with the following formula: This is the second objective function; Represents the decision variable for the p-th candidate edge node; This represents the total number of candidate edge nodes that were selected as actual deployed edge nodes. This represents the global average load, obtained by comparing the total load generated by the grid of all actually deployed edge nodes across the entire mine with the total number of actually deployed nodes. This indicates the actual load borne by the selected nth edge node candidate point; This represents an extremely small positive number, which is 0.001 in this embodiment. In this application, a load-sharing strategy within the same grid is adopted, meaning that if candidate points... Belongs to the grid and ,but , For grid The set of all candidate edge nodes within the region. This represents the total load that the g-th grid needs to process, which is the sum of the amount of data generated by all sensors in the grid per unit time; Let represent the decision variable for the k-th edge node candidate point.

[0043] The goal is to minimize the load imbalance among all selected edge nodes for actual deployment. By calculating the sum of the absolute deviations between the actual load borne by each selected candidate edge node and the global average load, the optimization algorithm is guided to achieve a uniform distribution of workload across the entire mine. This is combined with a load allocation mechanism: when the total load within a certain grid... Exceeding the load limit of a single device At that time, the optimization model will force the selection of multiple edge node candidate points within the grid (i.e., increase the number of edge node candidate points within the grid). The number of nodes (i.e., the total load of the grid is shared among multiple nodes, ensuring that no single edge node overloads and crashes, thereby improving the overall resilience and stability of the system; load limit) The throughput threshold is obtained by setting the processing power of a single edge node's hardware CPU.

[0044] The first constraint (covering constraint): ;in, For the actual mesh set where edge nodes are deployed, For set Grid in This constraint is used to ensure that at least one candidate edge node is actually selected within each selected grid. This ensures that there are no blind spots in computing and communication coverage in key areas of the mine.

[0045] The second constraint (load limit constraint): ; This is the maximum load supported by a single edge node hardware device; this constraint is used to ensure that the workload of any selected actual edge node never exceeds the physical limits of its hardware, preventing single-point overload and downtime.

[0046] The objective function and constraints are taken as input, and the population size (range is...) is set. In this embodiment, the value is 200. The number of generations is set (range [50, 500], set to 100 in this embodiment), and the crossover probability is set (range [0.7, 0.9], set to 0.8 in this embodiment). The NSGA-II algorithm is used to iteratively solve the multi-objective optimization problem of edge node candidate points, obtaining the Pareto optimal solution set. The solution set is then analyzed... and Normalize the solution, calculate the distance from each solution to the ideal point (0,0), and select the solution with the smallest distance as the final set of edge node deployment locations.

[0047] Based on the same inventive concept as the above method, this application embodiment also provides a lightweight deployment system for fog computing edge nodes in mine safety, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described methods for lightweight deployment of fog computing edge nodes in mine safety.

Claims

1. A lightweight deployment method for edge nodes in mine safety fog computing, characterized in that, The method includes the following steps: Collect 3D point cloud data of the mine to build a mine mesh model and mark the sensors and fog nodes already deployed in the mine; Analyze the sensor distribution density of each grid relative to its neighboring grids and the global grid to obtain the sensor distribution characteristics of each grid; within the preset maximum communication radius of the edge node device, analyze the distance characteristics between each grid and the grid where the sensor is located and the grid where the fog node is located, and combine the sensor distribution characteristics to calculate the edge node deployment adaptability of each grid. Based on the differences in edge node deployment fitness and location among the grids, all grids are clustered. Based on the overall distribution characteristics of edge node deployment fitness and the number of grids in each cluster, the grids with deployed edge nodes are initially labeled. The sensor coverage similarity of labeled grids among different clusters is analyzed to determine the actual set of grids with deployed edge nodes. Within the grid set of actual deployed edge nodes, candidate edge nodes are preset. With the optimization objectives of minimizing the total number of edge node deployments and minimizing the load imbalance between nodes, a multi-objective optimization model is constructed to optimize the deployment of candidate edge nodes, resulting in the final set of edge node deployment locations.

2. The lightweight deployment method for edge nodes of fog computing in mine safety as described in claim 1, characterized in that, The acquisition of sensor distribution characteristics for each grid specifically involves: Obtain the normalized result of the number of sensors in each grid under all its neighboring grids, and obtain the ratio of the number of sensors in each grid and its neighboring grids to the total number of sensors; multiply the obtained normalized result and the ratio to obtain the first local feature; The sensor distribution characteristics of each grid are obtained by averaging the first local features obtained with the normalized results of the number of sensors in each grid across all grids.

3. The lightweight deployment method for edge nodes of fog computing in mine safety as described in claim 1, characterized in that, The calculation of the edge node placement fitness of each grid is specifically as follows: For the i-th grid L(i), with the geometric center of L(i) as the reference, within the maximum communication radius of the edge node device, the maximum connected path distance between L(i) and all the grids where the sensors are located within this range is used as the data source access loss radius of L(i); the minimum connected path distance between L(i) and the grid where the nearest deployed fog node is located within this range is used as the uplink backhaul distance of L(i). Obtain the negative correlation mapping result of the sum of the uplink backhaul distance and the data source access loss radius of L(i), and perform positive fusion with the uplink backhaul distance and the sensor distribution characteristics to obtain the edge node deployment fitness of L(i).

4. The lightweight deployment method for edge nodes of fog computing in mine safety as described in claim 1, characterized in that, The distance metric used to cluster all grids is: The normalized geometric center coordinates of each grid and the fitness of edge node placement are concatenated to construct a four-dimensional feature vector, and the distance of the connected paths between the four-dimensional feature vectors is used as the clustering metric.

5. A lightweight deployment method for edge nodes in mine safety fog computing as described in claim 1, characterized in that, The initial marking of the mesh for deploying edge nodes is specifically as follows: Clusters are filtered based on the overall distribution of edge node deployment fitness across all grids. Within each filtered cluster, a representative grid is established based on the maximum edge node deployment fitness. Simultaneously, intra-cluster coverage deduplication is performed using the effective communication radius of the edge node devices. Finally, a grid set with preliminarily labeled edge node deployments is output. The effective communication radius is obtained by multiplying the maximum communication radius of the edge node devices by a preset reduction factor.

6. A lightweight deployment method for edge nodes of fog computing in mine safety as described in claim 1, characterized in that, The determination of the grid set for the actual deployment of edge nodes specifically includes: Calculate the similarity of sensors that can communicate with each other within the grid that are paired and marked as deployment edge nodes; use the mean of the sensor similarity of all pairs of grids marked as deployment edge nodes as the first judgment threshold. For two grids whose spatial distance is less than the maximum communication radius of the edge node device and whose sensor similarity is greater than the first judgment threshold, retain the grid with the highest edge node deployment fitness and remove the other grid. After traversing all combinations, the set of all the retained grids is taken as the grid set for actual deployment of edge nodes.

7. A lightweight deployment method for edge nodes of fog computing in mine safety as described in claim 1, characterized in that, The objective function of the multi-objective optimization model is specifically: Each candidate edge node in the actual deployed edge node grid is assigned a decision variable. When the first When a candidate point is actually selected for deployment of an edge node ;otherwise ; Minimize the total number of deployments: ;in, This refers to the set of all candidate edge nodes generated within all meshes marked as deployable edge nodes. Used to achieve lightweight deployment; Minimize load imbalance: ;in, To prevent extremely small positive numbers with a denominator of zero; For global average load, The actual load borne by the selected nth edge node candidate point; Used to minimize load imbalance between edge nodes.

8. A lightweight deployment method for edge nodes in mine safety fog computing as described in claim 7, characterized in that, A load-sharing strategy within the same grid is adopted, i.e., if candidate points Belongs to the grid and ,but , For grid The set of all candidate edge nodes within the region. This represents the total load that needs to be processed in the g-th grid. Let represent the decision variable for the k-th edge node candidate point.

9. A lightweight deployment method for edge nodes in mine safety fog computing as described in claim 7, characterized in that, The specific constraints for all multi-objective optimization models are as follows: First constraint: ;in, For the actual mesh set where edge nodes are deployed, For set Grid in ; For grid The set of all candidate edge nodes within the region; The second constraint: ;in, The maximum load supported by a single edge node hardware device.

10. A lightweight deployment system for fog computing edge nodes in mine safety, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-9.

Images