Resource configuration method and system for underground coal mine 5g communication

Abstract

The application discloses a resource configuration method and system for underground coal mine 5G communication, relates to the technical field of communication information processing, and comprises the following steps: acquiring data information of an underground coal mine roadway, combining a synchronous positioning and map construction algorithm and a wireless fingerprint matching algorithm, establishing a digital twin communication model, generating a channel profile according to real-time data, and extracting channel data of each spatial node; constructing a computing resource pool by using a 5G base station and an edge computing node, and encapsulating different service types into a plurality of network slice containers; according to the channel data, initially allocating each container by using a resource arranger, dynamically migrating tasks between different MEC nodes, and obtaining a resource allocation state; dividing a scheduling time according to the channel profile and the resource allocation state, and dynamically adjusting the length proportion and bandwidth occupation of each window according to the task migration state; and the application solves the problems of 5G communication coverage and interference in the underground coal mine through digital twinning and dynamic scheduling.

Description

Technical Field

[0001] This invention relates to the field of communication information processing technology, and more specifically, to a resource allocation method and system for 5G communication in underground coal mines. Background Technology

[0002] The underground communication environment in coal mines is highly complex and unique, making it a crucial research area for wireless communication and intelligent management. Traditional underground coal mine communication primarily relies on wired networks or low-speed wireless systems such as Wi-Fi or LTE. However, these technologies face challenges in the underground environment, including limited coverage, unstable links, high latency, and severe interference. The complex and variable underground roadways, coupled with multipath effects, rapid signal attenuation, metal equipment obstruction, and ore body vibrations, result in highly nonlinear and uncertain wireless signal propagation. In this environment, communication systems relying solely on empirical models or static deployments struggle to meet the demands of modern coal mines for data transmission speed, reliability, and quality of service. They also fail to support the high-bandwidth, low-latency communication needs of diverse operations such as real-time monitoring, intelligent scheduling, automated mining equipment, and unmanned mining vehicles.

[0003] For example, the invention patent with publication number CN119012392A discloses a resource allocation method for digital twin-assisted collaborative perception and edge collaboration, belonging to the field of mobile communication technology. It includes the following steps: S1: In a vehicle-to-everything (V2X) scenario, a multi-layer collaborative perception global map framework is proposed to help connected autonomous vehicles (CAVs) acquire global map perception results; S2: A digital twin (DT)-assisted edge collaboration method is proposed to provide offloading services for computationally intensive tasks; S3: The system task execution latency is analyzed based on the collaborative perception process; S4: A resource allocation scheme with the goal of minimizing the collaborative perception task completion latency is constructed, addressing a joint problem of bandwidth allocation, offloading, and computational resource allocation; S5: A multi-agent deep deterministic policy gradient (HAS-MADDPG) algorithm with a hybrid action space is proposed to solve the optimization problem. This method can optimize the resource allocation strategy for collaborative perception and edge offloading in V2X, reducing the execution latency of collaborative perception.

[0004] For example, the invention patent with publication number CN119697703A discloses a method for allocating wireless computing network resources based on digital twins. This method includes: constructing a wireless computing network architecture composed of a physical entity layer and its corresponding digital twin layer; the physical entity layer includes computing nodes composed of edge servers and terminal devices; fitting the state information of the edge servers and terminal devices through the digital twin of the digital twin layer; establishing an optimization problem with the goal of minimizing the total latency and total energy consumption of the wireless computing network architecture based on the fitted state information and the computing tasks of the wireless computing network architecture; and solving the optimization problem to obtain the optimal allocation strategy for computing resources. By acquiring the entity information of the computing nodes in real time through digital twins, the available computing resources and bandwidth of the computing nodes in the next time slot can be estimated, enabling the allocation of computing resources and task migration decisions while minimizing the total latency and total energy consumption of the wireless computing network architecture.

[0005] The above-disclosed technical solutions have at least the following technical problems:

[0006] Traditional 5G communication resource allocation technologies in coal mines typically lack the ability to dynamically identify and adaptively correct local interference. Communication links are easily affected by interference in complex underground environments, resulting in high data packet loss rates and unstable links. Furthermore, traditional scheduling schemes are mostly fixed-duration windows and static bandwidth allocations, which are difficult to cope with channel fluctuations and load changes caused by task migration in underground communication environments.

[0007] To address the above problems, this invention proposes a solution. Summary of the Invention

[0008] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a resource allocation method and system for 5G communication in underground coal mines, which solves the coverage and interference problems of 5G communication in underground coal mines through digital twins and dynamic scheduling.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A resource allocation method for 5G communication in underground coal mines includes: acquiring data information from underground coal mine roadways; establishing a digital twin communication model by combining synchronous positioning and mapping algorithms and wireless fingerprint matching algorithms; generating a channel profile map based on real-time data and extracting channel data from each spatial node; constructing a computing resource pool by combining 5G base stations and edge computing nodes based on the channel profile map, and encapsulating different service types into several network slice containers; initially allocating each container based on the channel data using a resource orchestrator, and dynamically migrating tasks between different MEC nodes to obtain the resource allocation status; dividing the scheduling time according to the channel profile and resource allocation status, and dynamically adjusting the duration ratio and bandwidth usage of each window according to the task migration status.

[0011] In a preferred embodiment, the acquisition of data information from underground coal mine roadways, combined with synchronous positioning and mapping algorithms and wireless fingerprint matching algorithms, establishes a digital twin communication model, and generates a channel profile map based on real-time data, extracting channel data from each spatial node, specifically as follows: Data information from underground coal mine roadways is acquired, and the data information undergoes spatiotemporal synchronization and coordinate self-calibration; the synchronized data information is input into a joint SLAM algorithm to reconstruct a three-dimensional underground spatial model of the coal mine roadways; uncertainty modeling of the pose information generated by SLAM is performed using a Bayesian graph optimization method, outputting the confidence weight of each spatial node to obtain a set of spatial nodes with confidence labels; several nodes are selected in the three-dimensional underground spatial model as signal sampling points, and the signal strength, channel state information, and multipath propagation parameters of each node are collected to form an initial channel sample set; through... Generative adversarial networks (GANs) expand the initial channel sample set to automatically generate virtual sample points and establish a channel fingerprint database. High-dimensional feature encoding is performed on the feature data in the channel fingerprint database to obtain a low-dimensional channel feature vector set. The three-dimensional downhole spatial model, spatial node confidence weights, and low-dimensional channel feature vectors are input into a graph neural network model to construct a digital twin communication model. The digital twin communication model is deployed on each MEC node, and combined with real-time acquired data and the twin model's prediction results, a predictive channel profile with a time dimension is generated. The predictive channel profile is analyzed to extract the instantaneous channel capacity, interference level, confidence level, and prediction bias of each spatial node, generating a channel dataset. Interference identification is performed on the channel dataset, abnormal channel states are detected based on cognitive radio principles, and local areas of the digital twin communication model are corrected through a self-attention mechanism.

[0012] In a preferred embodiment, the step of correcting the local region of the digital twin communication model through a self-attention mechanism is as follows: Feature extraction is performed on the channel dataset to construct a channel interference feature vector. Based on spectral feature analysis, the channel occupancy rate and interference signal power spectral density of each spatial node are calculated to obtain the interference intensity distribution in the spatial dimension. Based on historical channel statistics and real-time power spectral distribution, an interference detection threshold is obtained through an adaptive threshold algorithm, and abnormal signals are identified for each node according to the interference intensity distribution. If the channel capacity decreases beyond the interference detection threshold, the node is marked as an interference node, and an interference label matrix is ​​generated. A Gaussian mixture model is then used to further refine the data. The interference label matrix is ​​used for classification to determine the type and location of interference sources. Based on the interference classification results, the locations of interference sources and their influence radii are mapped to the spatial node indices of the digital twin communication model to obtain the set of nodes affected by interference. The predicted channel data of the set of nodes affected by interference is extracted and compared with the measured channel data to output the prediction deviation. The prediction deviations of the set of nodes affected by interference are arranged by node index to form the residual information matrix of the twin model. The residual information matrix is ​​input into a multi-head attention structure to output the weight changes of each spatial region. Based on the weight change results, the local region of the model most affected by interference is identified, the range of nodes that need to be corrected is determined, and the digital twin communication model is corrected.

[0013] In a preferred embodiment, the modification of the digital twin communication model is specifically as follows: Based on the required node range for modification, corresponding nodes and their adjacency relationships are extracted from the digital twin communication model to construct a local subgraph; the prediction bias of the local nodes is normalized to generate a residual vector, and the local subgraph and residual vector are input into a graph neural network to calculate the node gradient; the weights of the local nodes are updated according to gradient descent to initially correct the channel prediction error, while keeping the weights of undisturbed nodes unchanged; a multi-head attention mechanism is introduced into the features of each node in the local subgraph, and the gradient update magnitude is weighted and adjusted according to the residual magnitude, so that the modification is more focused on high-error nodes, resulting in the modified digital twin communication model; the channel state of the local nodes is re-predicted based on the modified digital twin communication model, and a new prediction error is calculated; if the prediction error is still higher than a preset threshold, the gradient calculation and attention weighting fine-tuning are repeated until the error converges or the maximum number of iterations is reached.

[0014] In a preferred embodiment, the step of constructing a computing resource pool for 5G base stations and edge computing nodes based on channel profiles, and encapsulating different service types into several network slice containers, specifically involves: evaluating the computing power, storage capacity, link latency, and bandwidth of 5G base stations and MEC nodes to form a node resource performance matrix; predicting the future communication load and interference trends of each node using a prediction model based on channel profiles and historical channel data; integrating the node resource performance matrix with the predicted communication load to form a preliminary computing resource pool; forming a service requirement feature table based on the bandwidth requirements, latency requirements, reliability requirements, and load of each service type; creating preliminary network slice containers for each service type in the computing resource pool, and providing constraints for each slice allocation based on the service requirement feature table; and dynamically weighting the resources within each slice using an intelligent optimization algorithm based on the constraints of the slice allocation.

[0015] In a preferred embodiment, the initial allocation of containers based on channel data using a resource orchestrator, and the dynamic task migration between different MEC nodes to obtain the resource allocation status, are as follows: Channel data and container service requirements are input into the computing resource pool, and container load, node resource status, and link quality are mapped to form a dynamic resource topology map; each MEC node is treated as an agent, generating an initial resource allocation strategy based on the dynamic resource topology map and slice service requirements; the initial resource allocation is executed, assigning slice containers to preferred MEC nodes, and recording the node binding, computing, storage, bandwidth usage, and link status of each container to form the initial resource allocation status. The system generates a resource allocation status table; based on channel profile data and historical load trends, it predicts the status of each MEC node and marks high-risk nodes; for high-risk nodes, it triggers a dynamic task migration strategy to migrate some slice container tasks to nearby MEC nodes with lower loads; during the task migration process, it dynamically scales up and down the migration containers according to the available computing, storage, and bandwidth resources of the target node, and updates the dynamic resource topology map in real time; through cognitive radio and link prediction algorithms, it optimizes the allocation of communication links between nodes in real time during the migration process; after the migration is completed, it updates the resource allocation status table, records the final resource occupancy, link status, and load of each slice container on the MEC node, and forms the latest resource allocation status.

[0016] In a preferred embodiment, the real-time optimization of inter-node communication link allocation during migration using cognitive radio and link prediction algorithms is as follows: A link prediction model is constructed to predict the link capacity and delay trends between the source node and the target node, and between the target node and its neighboring nodes, based on historical link status data and current channel data; the link congestion probability and transmission delay risk are calculated based on the prediction results, and data transmission links affecting the migration task are marked; spectrum sensing is performed on the marked links using cognitive radio technology, and available channels are identified to allocate optimal link frequency bands for the migration task; based on the link prediction model and cognitive radio analysis results, the optimal data transmission path for the migration task between nodes is determined, including a priority link and a backup link; the link status is monitored in real time during the migration process, and if link congestion, increased interference, or decreased signal-to-noise ratio is detected, the system switches to the backup link.

[0017] In a preferred embodiment, the step of dividing the scheduling time according to the channel profile and resource allocation status, and dynamically adjusting the duration ratio and bandwidth occupancy of each window according to the task migration status, is as follows: An initial scheduling time window is defined, and an initial duration ratio and bandwidth occupancy are allocated according to the service type to form a preliminary scheduling baseline scheme; the channel profile, resource allocation status, and task migration status are combined with the preliminary scheduling baseline scheme to form a resource utilization matrix; high-risk windows are marked according to the resource utilization matrix; the duration ratio and bandwidth occupancy of each window are dynamically adjusted using a reinforcement learning method based on service priority, QoS constraints, and high-risk window information; the channel capacity and interference trend of each window are predicted using a link prediction model, and the duration and bandwidth allocation of high-risk windows are adjusted in advance; the duration ratio and bandwidth allocation of each window are dynamically corrected according to link status, node load, and task migration status.

[0018] The system for resource allocation in 5G communication in underground coal mines includes a data acquisition module, a slicing module, an initial allocation module, and a dynamic adjustment module, with interconnections between the modules. The data acquisition module acquires data information from underground coal mine roadways, combines synchronous positioning and mapping algorithms with wireless fingerprint matching algorithms to establish a digital twin communication model, and generates a channel profile based on real-time data, extracting channel data from each spatial node. The slicing module constructs a computing resource pool from 5G base stations and edge computing nodes based on the channel profile and encapsulates different service types into several network slice containers. The initial allocation module performs initial allocation of each container based on channel data using a resource orchestrator and performs dynamic task migration between different MEC nodes to obtain the resource allocation status. The dynamic adjustment module divides the scheduling time according to the channel profile and resource allocation status, and dynamically adjusts the duration ratio and bandwidth usage of each window based on the task migration status.

[0019] The technical effects and advantages of the resource allocation method and system for 5G communication in underground coal mines according to the present invention are as follows:

[0020] 1. This invention acquires data from underground coal mine roadways and, combined with Simultaneous Localization and Mapping (SLAM) and wireless fingerprint matching algorithms, establishes a high-precision digital twin communication model, achieving accurate modeling of the complex three-dimensional underground space. By employing Bayesian graph optimization to model the uncertainty of location information and adding confidence labels to the model, the reliability of spatial node data and the accuracy of channel prediction are effectively improved. Self-supervised learning is used to expand the initial channel samples, generating virtual channel samples, and high-dimensional feature encoding is used to obtain low-dimensional channel feature vectors, further improving the channel fingerprint database of the digital twin communication model and providing comprehensive and accurate channel data support for subsequent resource allocation and scheduling.

[0021] 2. This invention utilizes a self-attention mechanism to correct local regions of a digital twin communication model, achieving real-time identification and compensation for channel interference. Specifically, it classifies interference signals and locates interference sources through spectral feature analysis and a Gaussian mixture model. Then, it calculates the weight changes of each spatial region using a multi-head attention structure, rapidly correcting the weights of nodes affected by interference. This significantly improves the model's prediction accuracy and robustness under abnormal channel conditions. This method effectively reduces the impact of interference on the system while ensuring communication quality, providing reliable protection for 5G communication in coal mines. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the resource allocation method for 5G communication in underground coal mines according to the present invention.

[0023] Figure 2 This is a schematic diagram of the system structure of the resource allocation method for 5G communication in underground coal mines according to the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Example 1, Figure 1 The present invention provides a resource allocation method for 5G communication in underground coal mines, comprising:

[0026] S1: Acquire data information from underground roadways in coal mines, combine synchronous positioning and map building algorithms and wireless fingerprint matching algorithms to establish a digital twin communication model, and generate channel profile maps based on real-time data to extract channel data from each spatial node;

[0027] In this embodiment, data information from underground coal mine roadways is acquired. A digital twin communication model is established by combining synchronous positioning and mapping algorithms with wireless fingerprint matching algorithms. A channel profile map is generated based on real-time data, and channel data for each spatial node is extracted, as detailed below:

[0028] Data information from underground coal mine roadways is acquired, including signal strength, channel status information, and terminal location (the terminal location is the three-dimensional spatial coordinates of the underground mobile communication node in the global coordinate system, which is obtained through the fusion positioning of inertial measurement unit, visual odometry, and wireless ranging).

[0029] The data information is spatiotemporally synchronized and self-calibrated, and a deep learning model based on spatiotemporal alignment network is used to realize automatic time synchronization and coordinate unification of multi-source data, and output multi-dimensional channel features after time synchronization.

[0030] The synchronized data information is input into the joint SLAM algorithm, which integrates visual features, laser point clouds and wireless signal strength information to reconstruct a three-dimensional underground space model of the coal mine roadway.

[0031] Uncertainty modeling of pose information generated by SLAM is performed using the Bayesian graph optimization method, and the confidence weight of each spatial node is output to obtain a set of spatial nodes with confidence labels.

[0032] Several nodes were selected as signal sampling points in the three-dimensional downhole space model, and the signal strength, channel state information and multipath propagation parameters of each node were collected to form an initial channel sample set.

[0033] By expanding the initial channel sample set through generative adversarial networks, virtual sample points are automatically generated to supplement blind zone data, forming a channel fingerprint database containing real and virtual samples. In the discriminator part of the GAN, confidence constraints are used to ensure that the spatial distribution of generated samples is consistent with the distribution of real channels, thereby improving the generalization ability in complex downhole environments.

[0034] High-dimensional feature encoding is performed on the feature data in the channel fingerprint database. Signal strength, channel state, multipath parameters and interference indicators are input into a convolutional autoencoder for feature compression to obtain a low-dimensional channel feature vector set.

[0035] The three-dimensional downhole spatial model, spatial node confidence weights, and low-dimensional channel feature vectors are input into the graph neural network model to establish a mapping between spatial topology and channel propagation characteristics, forming a digital twin communication model. The spatial node confidence weights refer to the set of spatial nodes with confidence labels.

[0036] The digital twin communication model is deployed on each MEC node to perform real-time simulation, state mapping and model synchronization of the underground communication status in coal mines;

[0037] Based on real-time collected data and the prediction results of the digital twin communication model, a predictive channel profile with a time dimension is generated through reinforcement learning and time series prediction algorithms. This profile reflects the channel capacity, interference intensity, and time-varying trends. The prediction results of the digital twin communication model include spatial topology state data (connection relationships between spatial nodes and reachability of each node to the base station or MEC node), channel propagation characteristic data (channel capacity of each spatial node), and channel confidence of each node.

[0038] The predictive channel profile is analyzed to extract the instantaneous channel capacity, interference, confidence level and prediction bias of each spatial node, and a channel dataset is generated.

[0039] Interference is identified in the channel dataset, abnormal channel states are detected based on cognitive radio principles, and local areas of the digital twin communication model are corrected through a self-attention mechanism.

[0040] In this embodiment, based on real-time collected data and the prediction results of the digital twin communication model, a predictive channel profile with a time dimension is generated using reinforcement learning and time series prediction algorithms, as detailed below:

[0041] The system acquires real-time channel data and digital twin communication model output data, and uses a time alignment algorithm to interpolate and synchronize data at different sampling frequencies to form a multi-dimensional time-series dataset arranged by time steps.

[0042] For each time step, the channel feature vector is normalized and encoded in high dimension, and then input into a bidirectional long short-term memory network (Bi-LSTM) to extract the sequential dependency features in the time series and obtain the channel state latent vector at the current time.

[0043] The channel state latent vector output by Bi-LSTM is input into the Transformer prediction model. The long-term dependencies between different time steps are captured through the self-attention mechanism, and the channel prediction feature sequence for multiple future time steps is output.

[0044] The channel prediction feature sequence is input into a reinforcement learning-driven temporal weighted policy network. The temporal weighted policy network takes the channel state vector and prediction error as input states, dynamically adjusts the prediction time step and feature dimension weights through the action space, and uses the satisfaction of the prediction error minimization constraint as the reward function to achieve adaptive weighting in the time dimension, thereby obtaining the optimized channel prediction feature sequence.

[0045] The optimized channel prediction feature sequence is restored to physical quantity data, which includes parameters such as channel capacity, interference intensity, signal-to-noise ratio and time delay jitter, and is reorganized into a multi-dimensional time series matrix according to the time axis order.

[0046] The multidimensional time series matrix is ​​normalized and interpolated for smoothing to generate a predictive channel profile with a time dimension. The profile reflects the dynamic trends of channel capacity, interference intensity, and signal-to-noise ratio over time.

[0047] In this embodiment, a self-attention mechanism is used to correct local regions of the digital twin communication model, as follows:

[0048] Feature extraction is performed on the channel dataset to construct a channel interference feature vector, which includes the channel capacity change rate, signal strength fluctuation amplitude, confidence decrease rate, and prediction bias.

[0049] Based on the channel interference feature vector, and using the spectral feature analysis method, the channel occupancy rate and interference signal power spectral density of each spatial node are calculated to obtain the interference intensity distribution in the spatial dimension.

[0050] Based on historical channel statistics and real-time power spectrum distribution, an interference detection threshold is obtained through an adaptive threshold algorithm. Abnormal signals are identified for each node according to the interference intensity distribution, distinguishing between normal and abnormal signals. The interference detection threshold is first obtained by extracting statistical features under normal conditions from historical channel capacity data and power spectrum data, calculating the baseline mean and fluctuation range of capacity change rate and power spectrum peak value, and then determining the initial threshold using the standard deviation method or quantile method. If the capacity decrease exceeds the lower limit of the baseline or the power spectrum peak value exceeds the upper limit of the baseline, it is judged as abnormal. Then, the statistical features and threshold are adaptively updated by combining real-time data through exponential weighting, so that the threshold can be dynamically adjusted with environmental changes. Finally, an interference detection threshold is formed to identify abnormal capacity decrease and abnormal power spectrum peak value.

[0051] Abnormal signals are identified for each spatial node. When the channel capacity drops below the interference detection threshold or an abnormal peak appears in the signal power spectrum, the node is marked as an interference node, and an interference tag matrix is ​​generated to record the interference identifier, interference intensity and corresponding timestamp of each node.

[0052] The interference label matrix is ​​classified using a Gaussian mixture model to determine the type of interference and the location of the interference source.

[0053] Based on the interference classification results, an interference impact area mapping model is established, which maps the location of the interference source and its impact radius to the spatial node index of the digital twin communication model, thus obtaining the interference impact node set;

[0054] Extract the predicted channel data of the node set affected by interference, compare it with the measured channel data, and output the prediction deviation;

[0055] The prediction bias of the node set affected by the interference is arranged by node index to form the twin model residual information matrix;

[0056] Input the residual information matrix into the multi-head attention structure and output the weight changes of each spatial region;

[0057] Based on the weight change results, identify the local areas of the model most affected by interference, determine the range of nodes that need to be corrected, and then correct the digital twin communication model.

[0058] In this embodiment, the digital twin communication model is modified as follows:

[0059] Based on the node range to be corrected as needed, extract the corresponding nodes and their adjacency relationships from the digital twin communication model to construct a local subgraph;

[0060] The prediction bias of local nodes is normalized to generate a residual vector, which is used to measure the node prediction error and serve as a correction constraint.

[0061] The local subgraph and residual vector are input into the graph neural network to calculate the node gradients. Backpropagation is performed only on the disturbed nodes and their neighbors.

[0062] The local node weights are updated based on gradient descent to initially correct the channel prediction error, while keeping the weights of undisturbed nodes unchanged.

[0063] A multi-head attention mechanism is introduced for the features of each node in the local subgraph. The gradient update magnitude is weighted and adjusted according to the residual size, so that the correction is more concentrated on the high error nodes, resulting in the corrected digital twin communication model.

[0064] The channel state of the local nodes is re-predicted based on the revised digital twin communication model, and the new prediction error is calculated.

[0065] If the prediction error is still higher than the preset threshold, repeat gradient calculation and attention-weighted fine-tuning until the error converges or the maximum number of iterations is reached. The preset threshold is determined by first using historical channel prediction results and corresponding measured data to statistically analyze the distribution characteristics of prediction errors of each node under normal conditions, and calculating their mean and fluctuation range. Then, by combining the validation set of the digital twin communication model under interference-free conditions, the prediction stability and convergence speed under different error thresholds are compared. The critical value that best distinguishes between acceptable errors and those that need further correction while ensuring the normal accuracy of the model is selected, and finally, this critical error is set as the preset threshold.

[0066] S2, based on the channel profile, construct a computing resource pool by combining 5G base stations and edge computing nodes, and encapsulate different service types into several network slice containers;

[0067] In this embodiment, a computing resource pool is constructed by combining 5G base stations and edge computing nodes based on the channel profile, and different service types are encapsulated into several network slice containers, as detailed below:

[0068] The computing power, storage capacity, link latency and bandwidth of 5G base stations and MEC nodes are evaluated to form a node resource performance matrix. Each element in the matrix corresponds to the available computing resources, storage resources, allocable bandwidth and link latency of the node.

[0069] Based on the channel profile and historical channel data, predictive models (such as LSTM or time series analysis) are used to predict the future communication load and interference trends of each node.

[0070] By integrating the node resource performance matrix with the predicted communication load, a preliminary computing resource pool is constructed, mapping the available bandwidth, computing and storage resources of each node to achieve unified management.

[0071] A service requirement characteristic table is formed based on the bandwidth requirements, latency requirements, reliability requirements, and load of the service type (such as eMBB, URLLC, mMTC).

[0072] In the computing resource pool, an initial network slice container is created for each service type, and constraints are provided for each slice according to the service requirement characteristic table, including bandwidth, upper limits of computing and storage resources, minimum quality of service guarantee, and fault tolerance redundancy settings.

[0073] The capacity, interference, and confidence level of each node in the channel profile are used as constraints for slice allocation. Node characteristics, service requirements, and predicted load are mapped to slice resource allocation. Intelligent optimization algorithms (such as genetic algorithms and particle swarm optimization) are used to dynamically weight resources within the slice to improve communication reliability and throughput. The dynamic weight allocation of resources within the slice involves adjusting bandwidth, computing, and storage resources according to priority, node load, and interference impact.

[0074] S3. Based on the channel data, the resource orchestrator performs initial allocation of each container and performs dynamic task migration between different MEC (edge ​​computing node) nodes to obtain the resource allocation status, which includes the slice resource allocation table and the task distribution status of each MEC node.

[0075] In this embodiment, based on channel data, the resource orchestrator performs initial allocation to each container and performs dynamic task migration between different MEC nodes to obtain the resource allocation status, as detailed below:

[0076] Channel data and container service requirements are input into the computing resource pool to map container load, node resource status, and link quality, forming a dynamic resource topology map.

[0077] Each MEC node is treated as an intelligent agent, generating an initial resource allocation strategy based on the dynamic resource topology and slice business requirements.

[0078] Perform initial resource allocation, assign slice containers to preferred MEC nodes, and record the node binding, compute / storage / bandwidth usage and link status of each container to form an initial resource allocation status table;

[0079] Based on channel profile data and historical load trends, the potential overload, delay, or interference at each MEC node is predicted, and high-risk nodes are marked.

[0080] For high-risk nodes, a dynamic task migration strategy is triggered to migrate some slice container tasks to nearby MEC nodes with lower load, while maintaining service QoS at a level no lower than the preset standard.

[0081] During the task migration process, the migration container is dynamically scaled up or down based on the available computing, storage, and bandwidth resources of the target node, and the dynamic resource topology is updated in real time.

[0082] By using cognitive radio and link prediction algorithms, the allocation of communication links between nodes is optimized in real time during the migration process to ensure that the data transmission latency of the migration task is minimized.

[0083] After the migration is complete, update the resource allocation status table to record the final resource usage, link status and load of each slice container on the MEC node, forming the latest resource allocation status.

[0084] In this embodiment, cognitive radio and link prediction algorithms are used to optimize the allocation of communication links between nodes in real time during the migration process, as detailed below:

[0085] A link prediction model is constructed to predict the link capacity and delay trends between the source node and the target node, as well as between the target node and its neighboring nodes, based on historical link status data and current channel data. The historical link status data includes the channel capacity, signal-to-noise ratio, interference, bit error rate, delay, and packet loss rate between the source node and the target node, and between the target node and its neighboring nodes.

[0086] The link prediction model captures the temporal correlation of link states through a Long Short-Term Memory (LSTM) network and models the spatial dependence between nodes through a multi-agent cooperation mechanism to achieve spatiotemporally coupled link prediction.

[0087] Calculate the link congestion probability and transmission delay risk based on the prediction results, and mark the data transmission links that affect the migration task;

[0088] Cognitive radio technology is used to perform spectrum sensing, interference analysis, and available channel identification on marked links, so as to allocate the optimal link frequency band for the migration task.

[0089] Based on the link prediction model and cognitive radio analysis results, the optimal data transmission path between nodes for the migration task is determined, including the priority link and the backup link;

[0090] During the migration process, the link status is monitored in real time. If link congestion, increased interference, or decreased signal-to-noise ratio is detected, the link is switched to a backup link or the frequency band is reallocated.

[0091] S4 divides the scheduling time according to the channel profile and resource allocation status, and dynamically adjusts the duration ratio and bandwidth usage of each window according to the task migration status.

[0092] In this embodiment, the scheduling time is divided according to the channel profile and resource allocation status, and the duration ratio and bandwidth usage of each window are dynamically adjusted according to the task migration status, as follows:

[0093] Define the initial scheduling time window, allocate the initial duration ratio and bandwidth usage according to the service type (such as eMBB, URLLC, mMTC), and form a preliminary scheduling baseline scheme;

[0094] By combining the channel profile, resource allocation status, and task migration status with the preliminary scheduling baseline scheme, the available bandwidth, node load, latency risk, and interference risk for each window are calculated to form a resource utilization matrix.

[0095] Based on the resource utilization matrix, high-risk windows are marked where bandwidth usage is too high, or latency or interference may exceed limits.

[0096] Based on service priority, QoS constraints, and high-risk window information, the duration ratio and bandwidth usage of each window are dynamically adjusted through reinforcement learning methods. The QoS constraints include bandwidth, latency, and packet loss rate.

[0097] By predicting the channel capacity and interference trends of each window through the link prediction model, the duration and bandwidth allocation of high-risk windows are adjusted in advance to ensure that the QoS requirements of migration tasks and high-priority services are met.

[0098] The duration ratio and bandwidth allocation of each window are dynamically adjusted based on link status, node load, and task migration status.

[0099] The duration ratio and bandwidth allocation of each window are dynamically adjusted based on link status, node load, and task migration status, as detailed below:

[0100] ;

[0101] ;

[0102] In the formula: It adjusts the duration ratio of each window. Indicates the link status. Indicates node load. Indicates the migration task status. , , , , and To adjust the weights, It corrects the bandwidth allocation for each window.

[0103] Example 2, Figure 2 The present invention provides a system for resource allocation method of 5G communication in underground coal mines, comprising a data acquisition module, a slicing module, an initial allocation module, and a dynamic adjustment module, with connections between the modules;

[0104] The data acquisition module is used to acquire data information from underground roadways in coal mines. It combines synchronous positioning and map building algorithms with wireless fingerprint matching algorithms to establish a digital twin communication model and generate channel profile maps based on real-time data, extracting channel data from each spatial node.

[0105] The slicing module is used to build a computing resource pool by combining 5G base stations and edge computing nodes according to the channel profile map, and to encapsulate different service types into several network slice containers.

[0106] The initial allocation module is used to perform initial allocation of each container based on channel data through the resource orchestrator, and to perform dynamic task migration between different MEC nodes to obtain the resource allocation status.

[0107] The dynamic adjustment module is used to divide the scheduling time according to the channel profile and resource allocation status, and dynamically adjust the duration ratio and bandwidth usage of each window according to the task migration status.

[0108] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0109] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0110] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0111] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0112] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0113] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A resource configuration method for underground coal mine 5G communication, characterized in that, The method comprises the following steps: Obtain data information of the underground roadway of the coal mine, combine the synchronous positioning and map construction algorithm and the wireless fingerprint matching algorithm, establish a digital twin communication model, and generate a channel profile based on real-time data to extract channel data of each spatial node; According to the channel profile, build a computing resource pool with 5G base stations and edge computing nodes, and encapsulate different business types into several network slice containers; According to the channel data, perform initial allocation of each container through a resource orchestrator, and perform dynamic task migration between different edge computing nodes to obtain a resource allocation state; According to the channel profile and the resource allocation state, divide the scheduling time, and dynamically adjust the time length ratio and bandwidth occupation of each window according to the task migration state; The method for obtaining data information of the underground roadway of the coal mine, combining the synchronous positioning and map construction algorithm and the wireless fingerprint matching algorithm, establishing a digital twin communication model, and generating a channel profile based on real-time data to extract channel data of each spatial node is as follows: Obtain data information of the underground roadway of the coal mine, and perform space-time synchronization and coordinate self-calibration on the data information; Input the synchronized data information into a joint SLAM algorithm to reconstruct a three-dimensional underground space model of the underground roadway of the coal mine; Perform uncertainty modeling on the pose information generated by the SLAM through a Bayesian graph optimization method, output the confidence weight of each spatial node, and obtain a spatial node set with a confidence label; Select several nodes in the three-dimensional underground space model as signal sampling points, and collect the signal strength, channel state information and multipath propagation parameters of each node to form an initial channel sample set; Expand the initial channel sample set through a generative adversarial network to automatically generate virtual sample points and establish a channel fingerprint database; Perform high-dimensional feature coding on the feature data in the channel fingerprint database to obtain a low-dimensional channel feature vector set; Input the three-dimensional underground space model, the spatial node confidence weight and the low-dimensional channel feature vector into a graph neural network model to construct a digital twin communication model; Deploy the digital twin communication model on each edge computing node, and generate a predicted channel profile with a time dimension based on the real-time collected data information and the prediction result of the twin model; Analyze the predicted channel profile to extract the instantaneous channel capacity, interference degree, confidence degree and prediction deviation of each spatial node to generate a channel data set; Perform interference identification on the channel data set, detect abnormal channel states based on the cognitive radio principle, and correct the local area of the digital twin communication model through a self-attention mechanism.

2. The resource configuration method for underground coal mine 5G communication of claim 1, characterized in that, The method for correcting the local area of the digital twin communication model through a self-attention mechanism is as follows: Perform feature extraction on the channel data set to construct a channel interference feature vector, calculate the channel occupancy rate and interference signal power spectral density of each spatial node based on a spectral feature analysis method, and obtain the interference intensity distribution in the spatial dimension; According to the historical channel statistical data and the real-time power spectral distribution, obtain an interference detection threshold value through an adaptive threshold algorithm, and identify abnormal signals of each node according to the interference intensity distribution. If the channel capacity drops more than the interference detection threshold, the node is marked as an interference node, and an interference label matrix is generated; Classify the interference label matrix using a Gaussian mixture model to determine the type of interference and the location of the interference source; According to the interference classification result, map the interference source position and its influence radius to the spatial node index of the digital twin communication model to obtain the interference affected node set; Extract the predicted channel data of the interference affected node set and compare it with the measured channel data to output the prediction deviation; Arrange the prediction deviation of the interference affected node set according to the node index to form a twin model residual information matrix; Input the residual information matrix into a multi-head attention structure to output the weight changes of each spatial region; According to the weight change result, identify the model local area most affected by the interference, determine the node range that needs to be corrected, and correct the digital twin communication model.

3. The resource configuration method for underground coal mine 5G communication of claim 2, characterized in that, The correction of the digital twin communication model is as follows: According to the node range that needs to be corrected, extract the corresponding nodes and their adjacency relationships from the digital twin communication model to construct a local subgraph; Normalize the prediction deviation of the local nodes to generate a residual vector, and input the local subgraph and residual vector into a graph neural network to calculate the node gradient; Update the local node weight according to the gradient descent to preliminarily correct the channel prediction error, while keeping the weight of the nodes not affected by the interference unchanged; Introduce a multi-head attention mechanism to the features of each node in the local subgraph, and weight adjust the gradient update amplitude according to the residual size to make the correction more concentrated on the high error nodes, obtaining the corrected digital twin communication model; According to the corrected digital twin communication model, re-predict the channel state of the local nodes and calculate the new prediction error; If the prediction error is still higher than the preset threshold, repeat the gradient calculation and attention weighted adjustment until the error converges or the maximum iteration number is reached.

4. The resource configuration method for underground coal mine 5G communication of claim 3, characterized in that, The 5G base station and edge computing node are constructed into a computing resource pool according to the channel profile, and different business types are packaged into several network slice containers as follows: Evaluate the computing power, storage capacity, link delay, and bandwidth of the 5G base station and edge computing node to form a node resource performance matrix; According to the channel profile and historical channel data, use a prediction model to predict the future communication load and interference trend of each node; Integrate the node resource performance matrix and the predicted communication load to form a preliminary computing resource pool; Form a business demand feature table according to the bandwidth demand, delay requirement, reliability requirement, and load of the business type; Create a preliminary network slice container for each business type in the computing resource pool, and assign constraint conditions to each slice according to the business demand feature table; According to the constraint conditions assigned to the slice, dynamically allocate weights to the resources within the slice through an intelligent optimization algorithm.

5. The resource configuration method for underground coal mine 5G communication of claim 4, characterized in that, According to the channel data, the resource orchestrator performs initial allocation of each container and dynamic task migration between different edge computing nodes to obtain the resource allocation state as follows: Input the channel data and container business demand into the computing resource pool to map the container load, node resource state, and link quality to form a dynamic resource topology graph; Each edge computing node is taken as an agent to generate an initial resource allocation strategy according to a dynamic resource topology graph and slice service requirements; An initial resource allocation is performed to allocate slice containers to preferred edge computing nodes, and node binding, computing, storage, bandwidth occupation and link state of each container are recorded to form an initial resource allocation state table; High-risk nodes are marked by predicting each edge computing node according to a channel profile graph and historical load trend; For the high-risk nodes, a dynamic task migration strategy is triggered to migrate part of the slice container tasks to adjacent edge computing nodes with lower load; In the task migration process, the migrated containers are dynamically scaled according to the available computing, storage and bandwidth resources of the target nodes, and the dynamic resource topology graph is updated in real time; The inter-node communication link allocation is optimized in real time during the migration process through cognitive radio and link prediction algorithms; After the migration is completed, the resource allocation state table is updated to record the final resource occupation, link state and load of each slice container on the edge computing nodes to form the latest resource allocation state.

6. The resource configuration method for underground coal mine 5G communication of claim 5, characterized in that, The inter-node communication link allocation is optimized in real time during the migration process through cognitive radio and link prediction algorithms, and the specific implementation is as follows: A link prediction model is constructed to predict the link capacity and delay trend between the source node and the target node and between the target node and the adjacent nodes according to historical link state data and current channel data; The link congestion probability and transmission delay risk are calculated according to the prediction results, and the data transmission links affecting the migrated tasks are marked; Through cognitive radio technology, the marked links are subjected to spectrum sensing, and available channels are identified to allocate the optimal link frequency band for the migrated tasks; According to the link prediction model and the cognitive radio analysis results, the optimal data transmission path between nodes for the migrated tasks is determined, and the transmission path includes a priority link and a backup link; The link state is monitored in real time during the migration process, and if link congestion, increased interference and decreased signal-to-noise ratio are detected, the backup link is switched to.

7. The resource configuration method for underground coal mine 5G communication of claim 6, characterized in that, The scheduling time is divided according to the channel profile graph and the resource allocation state, and the length proportion and bandwidth occupation of each window are dynamically adjusted according to the task migration state, and the specific implementation is as follows: An initial scheduling time window is defined, the initial length proportion and bandwidth occupation are allocated according to the business type, and a preliminary scheduling reference scheme is formed; The channel profile graph, resource allocation state and task migration state are combined with the preliminary scheduling reference scheme to form a resource utilization matrix; High-risk windows are marked according to the resource utilization matrix; The length proportion and bandwidth occupation of each window are dynamically adjusted according to the business priority, QoS constraint and high-risk window information through reinforcement learning method; The channel capacity and interference trend of each window are predicted through the link prediction model, and the length and bandwidth allocation of the high-risk window are adjusted in advance; The length proportion and bandwidth allocation of each window are dynamically corrected according to the link state, node load and task migration state.

8. A system for using the resource configuration method for underground coal mine 5G communication according to any one of claims 1-7, characterized in that, The system comprises a data collection module, a slice module, an initial allocation module and a dynamic adjustment module, and the modules are connected. The data acquisition module is configured to acquire data information of a coal mine underground roadway, establish a digital twin communication model by combining a synchronous positioning and map construction algorithm and a wireless fingerprint matching algorithm, and generate a channel profile according to real-time data to extract channel data of each spatial node. The slicing module is configured to construct a computing resource pool by the 5G base station and the edge computing node according to the channel profile, and encapsulate different service types into a plurality of network slice containers. The initial allocation module is configured to perform initial allocation on each container by a resource orchestrator according to the channel data, and perform dynamic task migration between different edge computing nodes to obtain a resource allocation state. The dynamic adjustment module is configured to divide a scheduling time according to the channel profile and the resource allocation state, and dynamically adjust a time length ratio and a bandwidth occupation of each window according to a task migration state.

Images