Network resource adaptive cooperative scheduling communication method

Abstract

The invention discloses a network resource self-adaptive cooperative scheduling communication method, which comprises the following steps of: 1, establishing a dynamic sensing layer taking user equipment (UE) as a center, and collecting multi-dimensional dynamic demand characteristic data of individual UE and a UE group in real time; 2, constructing a demand prediction model based on historical data and real-time feature data, predicting resource demand change trends of individual UE and UE groups in a future time window, and generating a demand prediction map; 3, according to the demand prediction map, a dynamic resource scheduling strategy is generated through a distributed cooperative scheduling engine, and the scheduling strategy acts on a base station RAN, an edge computing node MEC, a core network control plane function and adjacent UE equipment at the same time; and step 4, executing a dynamic resource scheduling strategy, synchronously implementing resource reconfiguration on multiple layers of a communication protocol stack, and completing joint collaborative allocation of base station spectrum resources, edge computing resources, core network bandwidth resources and adjacent UE relay resources.

Description

Technical Field

[0001] This invention relates to the field of network communication technology, and in particular to a network resource adaptive collaborative scheduling communication method. Background Technology

[0002] With the rapid development of mobile internet and the Internet of Things, the scale of network users and the types of services are experiencing explosive growth, highlighting the increasing demand for dynamic, differentiated, and collaborative communication resources. In traditional network architectures, resource scheduling is typically centered on network devices, with base stations, core networks, and other infrastructure independently executing static or semi-static resource allocation strategies. In this model, resource management units (such as spectrum, computing, and bandwidth) are dispersed across different levels (access network, edge network, and core network), and scheduling decisions are mainly based on local state information (such as base station load and link quality), making it difficult to fully perceive the real-time dynamic demand characteristics of individual user equipment (UE) and groups (such as service type switching, changes in movement trajectories, and fluctuations in device status).

[0003] The main drawbacks of existing technologies include: resource allocation strategies rely on preset rules or historical averages, which cannot capture user dynamic behavior in real time, leading to a mismatch between resource supply and actual demand, resulting in idle resources or service degradation; resource pools among base stations, edge nodes, core networks, and user equipment are fragmented, lacking a unified collaborative scheduling framework, making it difficult to jointly optimize heterogeneous resources such as spectrum, computing, and relays; lack of accurate prediction capabilities for the future resource needs of individual users / groups, making it impossible to reserve resources in advance to cope with peak loads or service migration, exacerbating service latency and resource fragmentation; resource reconfiguration is usually implemented at a single protocol layer (such as the physical layer), failing to achieve synchronous dynamic adjustment across layers (physical layer / MAC layer / network layer), which limits end-to-end performance improvement. Summary of the Invention

[0004] Purpose of the invention: The technical problem to be solved by the present invention is to provide a network resource adaptive collaborative scheduling communication method to address the shortcomings of the prior art.

[0005] To address the aforementioned technical problems, this invention discloses a network resource adaptive cooperative scheduling communication method, comprising the following steps:

[0006] Step 1: Establish a dynamic perception layer centered on the user equipment (UE) to collect multi-dimensional dynamic demand characteristic data of individual UEs and UE groups in real time;

[0007] Step 2: Construct a demand forecasting model based on historical data and real-time feature data to predict the changing trends of resource demand for individual UEs and UE groups within future time windows, and generate a demand forecasting map;

[0008] Step 3: Based on the demand forecast map, a dynamic resource scheduling strategy is generated through a distributed collaborative scheduling engine. The scheduling strategy applies simultaneously to the base station RAN, edge computing node MEC, core network control plane functions, and adjacent UE devices.

[0009] Step 4: Execute the dynamic resource scheduling strategy, implement resource reconfiguration synchronously at multiple layers of the communication protocol stack, and complete the joint and coordinated allocation of base station spectrum resources, edge computing resources, core network bandwidth resources, and adjacent UE relay resources.

[0010] The multidimensional dynamic demand characteristics mentioned in step 1 include service type characteristics, real-time mobility characteristics, device status characteristics, and channel quality indicators.

[0011] The collection of dynamic demand characteristics in step 1 includes: acquiring device status characteristics, including battery capacity, processor load, and screen activation status, through sensors built into the user equipment (UE); for example, a battery power meter can be used to acquire battery capacity, a CPU utilization monitor can be used to acquire processor load, and a screen driver chip or system power management module can be used to acquire screen activation status; acquiring mobility characteristics, including the UE's own movement speed, movement trajectory prediction, and cell handover frequency, through base station measurement reports; identifying service type characteristics, including classification labels for latency-sensitive services, bandwidth-intensive services, and computationally intensive services, through deep packet inspection (DPI); and acquiring instantaneous channel quality indications through channel state information (CSI) feedback.

[0012] The demand forecasting model described in step 2 adopts a spatiotemporal joint modeling approach: for individual UEs, an LSTM-based time-series forecasting model is used, which takes historical resource usage sequences and real-time feature vectors as inputs and outputs the bandwidth demand, computing resource demand, and maximum tolerable latency for the next N time slots; for the UE group, a graph neural network (GNN) model is used, which constructs a dynamic topology graph of geographically adjacent UEs and predicts the peak resource demand of the group and the distribution of regional hotspots by aggregating node features.

[0013] The demand forecasting model described in step 2 includes an elastic resource reservation mechanism. Based on the peak resource demand identified in the demand forecasting map, resources are dynamically reserved in base stations, edge nodes, and the core network according to the reservation ratio to cope with the predicted peak resource demand and improve service stability. The formula for calculating the reservation ratio P is: P = k × predicted peak utilization rate + (1-k) × historical maximum utilization rate, where k is the prediction confidence weight, with a value range of 0.6-0.9. When the actual resource utilization rate is lower than 80% of the reservation threshold for n consecutive time slots, the reserved resources are automatically released to the public resource pool.

[0014] The generation of the dynamic resource scheduling strategy described in step 3 includes a joint optimization objective, defined as: minimizing total system overhead = α × transmission delay + β × energy consumption + γ × resource fragmentation rate; where α, β, and γ are weighting coefficients, all ranging from [0,1], and α + β + γ = 1, used to balance the relative importance of each optimization objective; constraints include: the number of available base stations (RBs), edge node computing capacity, core network maximum bandwidth, and upper limit of relay capability of adjacent UEs; the optimal resource allocation scheme is solved using an improved distributed Q-learning algorithm.

[0015] The improved distributed Q-learning algorithm includes a conflict resolution mechanism to resolve resource allocation decision conflicts among multiple distributed scheduling nodes. When resource allocation schemes of multiple scheduling nodes (base station / MEC / core network) conflict, a three-level arbitration process is executed, proceeding sequentially:

[0016] The three-tier arbitration process includes:

[0017] Local arbitration: The edge node adjusts resource allocation according to the UE service priority label [eMBB > uRLLC > mMTC], where eMBB is the enhanced mobile broadband label, uRLLC is the ultra-reliable low-latency communication label, and mMTC is the massive machine-type communication label.

[0018] Regional arbitration: The core network SDN controller re-plans cross-domain routing paths to prioritize latency-sensitive services;

[0019] Global arbitration: The cloud management platform coordinates the migration of resources from multiple edge nodes to balance regional load differences.

[0020] Local arbitration is the first to be executed. If local arbitration fails to resolve the conflict, it is escalated to regional arbitration. If regional arbitration also fails to resolve the conflict, it is escalated to global arbitration.

[0021] The three-level arbitration mechanism resolves resource allocation decision conflicts among multiple nodes in distributed scheduling. Through tiered processing, it prioritizes resolving conflicts locally to reduce communication overhead. When local resolution fails, conflicts are escalated to the regional or global level to ensure resource requirements for high-priority services (such as uRLLC), preventing resource allocation deadlocks and improving overall system efficiency and stability. Specifically, the three-level arbitration mechanism rapidly adjusts resource allocation locally using service priority labels (eMBB > uRLLC > mMTC) to reduce latency; at the regional level, it re-plans routes through the SDN controller to ensure the continuity of cross-domain services; and at the global level, it balances load through resource migration to prevent local overload, thereby significantly reducing service interruption rates and service quality degradation caused by resource conflicts.

[0022] The collaborative scheduling of adjacent UE device resources corresponding to the execution of the dynamic resource scheduling strategy in step 4 specifically includes: establishing a device-to-device (D2D) collaborative resource pool between UEs, and including adjacent UEs that meet the signal strength threshold in the resource pool; when it is predicted that the target UE will enter a high load state, selecting a relay UE from the resource pool and dynamically allocating its idle antenna resources, buffer resources, and computing resources to the target UE; resource allocation is achieved through a pre-configured D2D link, and the core network authorizes resource usage rights through control plane signaling.

[0023] The communication protocol stack described in step 4 consists of three layers: physical layer, MAC layer, and network layer.

[0024] Beneficial effects:

[0025] This invention achieves global elastic adaptation and efficient cross-domain integration of network resources by constructing a dynamic collaborative scheduling framework centered on user equipment. Based on real-time perception of multi-dimensional features and a spatiotemporal joint prediction model, it anticipates changes in individual and group resource demands, fundamentally solving the mismatch between resource supply and actual demand. Through a distributed scheduling engine, it jointly schedules base station spectrum, edge computing, core network bandwidth, and relay resources of adjacent UEs, breaking the traditional fragmented resource pool situation and reducing transmission latency while minimizing energy consumption. Relying on a multi-layer protocol stack synchronous reconfiguration mechanism and elastic resource reservation strategy, it effectively copes with sudden load peaks, reducing the service quality default rate of latency-sensitive services and lowering resource fragmentation. An improved Q-learning algorithm combined with a three-level arbitration mechanism (local / regional / global) dynamically resolves resource allocation conflicts among multiple nodes, ensuring the latency stability of high-priority services. The three-level arbitration mechanism prioritizes resolving conflicts locally to reduce communication overhead; when local resolution fails, it escalates to the regional or global level to ensure the resource needs of high-priority services (such as uRLLC), avoiding resource allocation deadlock and improving overall system efficiency and stability. Attached Figure Description

[0026] Figure 1 This is a flowchart of the algorithm of the present invention. Detailed Implementation

[0027] This embodiment describes a scenario in a high-density user environment (100 UEs / km²) within a smart park, encompassing mobile inspection robots (uRLLC service), 4K video surveillance terminals (eMBB service), and environmental sensors (mMTC service). To verify the effectiveness of this invention, the method employed (referred to as the "cooperative scheduling scheme") will be compared with a traditional static resource allocation scheme (referred to as the "comparison scheme").

[0028] The execution process of this embodiment is as follows:

[0029] Step 1: Establish a dynamic perception layer centered on the user equipment (UE) to collect multi-dimensional dynamic demand characteristic data of individual UEs and UE groups in real time;

[0030] Step 2: Construct a demand forecasting model based on historical data and real-time feature data to predict the changing trends of resource demand for individual UEs and UE groups within future time windows, and generate a demand forecasting map;

[0031] Step 3: Based on the demand forecast map, a dynamic resource scheduling strategy is generated through a distributed collaborative scheduling engine. The scheduling strategy applies simultaneously to the base station RAN, edge computing node MEC, core network control plane functions, and adjacent UE devices.

[0032] Step 4: Execute the dynamic resource scheduling strategy, implement resource reconfiguration synchronously at multiple layers of the communication protocol stack, and complete the joint and coordinated allocation of base station spectrum resources, edge computing resources, core network bandwidth resources, and adjacent UE relay resources.

[0033] The collection of dynamic demand characteristics in step 1 includes: acquiring device status characteristics, including battery capacity, processor load, and screen activation status, through sensors built into the user equipment (UE); acquiring mobility characteristics, including the UE's movement speed, movement trajectory prediction, and cell handover frequency, through base station measurement reports; identifying service type characteristics, including classification labels for latency-sensitive services, bandwidth-intensive services, and computationally intensive services, through deep packet inspection (DPI); and acquiring instantaneous channel quality indications through channel state information (CSI) feedback.

[0034] (I) In this embodiment, the dynamic feature acquisition in step 1 is specifically as follows:

[0035] ① Equipment status acquisition: The UE1 inspection robot reports the equipment status (battery capacity 35%, processor load 82%, screen continuously active) in real time through built-in sensors (such as battery power meter and CPU utilization monitor).

[0036] ② Mobility monitoring: The base station analyzes the motion characteristics of UE1 based on the measurement report (speed 8m / s, predicted to enter the weak coverage area in 10 seconds, handover frequency 3 times / minute).

[0037] ③ Service type identification: Edge node MEC1 marks the traffic of video terminal UE2 as bandwidth-intensive service (eMBB) through deep packet inspection.

[0038] ④ Channel quality assessment: The core network integrates CSI feedback and determines that the channel quality between UE1 and the base station is level C (RSRP=-105dBm).

[0039] The demand forecasting model described in step 2 adopts a spatiotemporal joint modeling approach: for individual UEs, an LSTM-based time-series forecasting model is used, which takes historical resource usage sequences and real-time feature vectors as inputs and outputs the bandwidth demand, computing resource demand, and maximum tolerable latency for the next N time slots; for the UE group, a graph neural network (GNN) model is used, which constructs a dynamic topology graph of geographically adjacent UEs and predicts the peak resource demand of the group and the distribution of regional hotspots by aggregating node features.

[0040] The demand forecasting model includes an elastic resource reservation mechanism. Based on the peak resource demand identified in the demand forecasting map, resources are dynamically reserved in base stations, edge nodes, and the core network according to the reservation ratio to cope with the predicted peak resource demand and improve service stability. The reservation ratio P is calculated as follows: P = k × predicted peak utilization rate + (1-k) × historical maximum utilization rate, where k is the prediction confidence weight, with a value range of 0.6-0.9. When the actual resource utilization rate is lower than 80% of the reservation threshold for n consecutive time slots, the reserved resources are automatically released to the public resource pool.

[0041] (ii) In this embodiment, step 2, demand forecasting and resource reservation, specifically involves:

[0042] Individual demand prediction: The LSTM model takes UE1's historical resource usage records and real-time characteristics as input and outputs the predicted values ​​for the next 5 time slots: ① Time slot 1: Bandwidth requirement 25Mbps, computational requirement 0.8TFLOPS, maximum tolerable latency 15m; ② Time slot 3: Bandwidth requirement increases to 50Mbps (redundancy requirement for weak coverage areas).

[0043] Hotspot prediction: The GNN model constructs a dynamic topology map for 20 UEs within a 50-meter radius. Prediction results: In the region with coordinates (X12, Y34), time slot 3 has a peak computing resource demand of 8 TFLOPS; the peak bandwidth in the northeast region reaches 1.2 Gbps.

[0044] Flexible resource reservation: Base stations reserve 80 resource blocks (RBs) according to the formula P=0.7×85%+0.3×70%=80.5%; edge nodes reserve 6.4 TFLOPS of computing resources (80% of peak demand) for hotspot areas. This preparatory scheduling provides critical resource guarantees for coping with peak loads.

[0045] The dynamic resource scheduling strategy generated in step 3 includes a joint optimization objective, defined as: minimizing total system overhead = α × transmission delay + β × energy consumption + γ × resource fragmentation rate; where α, β, and γ are weighting coefficients, all ranging from [0,1], and α + β + γ = 1, used to balance the relative importance of each optimization objective; constraints include: the number of available base stations (RBs), edge node computing capacity, core network maximum bandwidth, and upper limit of adjacent UE relay capacity; the optimal resource allocation scheme is solved using an improved distributed Q-learning algorithm to resolve resource allocation decision conflicts among multiple distributed scheduling nodes. When resource allocation schemes of multiple scheduling nodes (base station / MEC / core network) conflict, a three-level arbitration process is executed. This is performed sequentially:

[0046] (III) In this embodiment, the dynamic scheduling and conflict resolution in step 3 are specifically...

[0047] Resource allocation conflict: Time slot 3 base station needs to allocate 40 RBs to UE1 to compensate for weak coverage, while MEC1 needs to preempt the same RB resources for video terminal UE2.

[0048] The three-level arbitration is executed as follows: (1) Local arbitration: MEC1 allocates RB to UE1 (inspection robot) based on service priority (uRLLC>eMBB); (2) Regional arbitration: the core network SDN controller replans the path for UE2 (video terminal): connecting to MEC2 via relay UE3; (3) Global arbitration: the cloud platform migrates the 2TFLOPS computing task of MEC1 to the low-load MEC3 node. This conflict resolution mechanism ensures the continuity of high-priority services. The three-level arbitration is executed in sequence, with local arbitration being executed first. If local arbitration cannot resolve the conflict, it is upgraded to regional arbitration. If regional arbitration still cannot resolve the conflict, it is upgraded to global arbitration. The three-level arbitration mechanism solves the problem of resource allocation decision conflicts among multiple nodes in distributed scheduling. Through hierarchical processing, conflicts are resolved locally first, reducing communication overhead. When local resolution is not possible, it is upgraded to the regional or global level to ensure the resource requirements of high-priority services (such as uRLLC), avoid resource allocation deadlock, and improve the overall efficiency and stability of the system.

[0049] The collaborative scheduling of adjacent UE device resources corresponding to the execution of the dynamic resource scheduling strategy in step 4 specifically includes: establishing a device-to-device (D2D) collaborative resource pool between UEs, and including adjacent UEs that meet the signal strength threshold in the resource pool; when it is predicted that the target UE will enter a high load state, selecting a relay UE from the resource pool and dynamically allocating its idle antenna resources, buffer resources, and computing resources to the target UE; resource allocation is achieved through a pre-configured D2D link, and the core network authorizes resource usage rights through control plane signaling.

[0050] (iv) In this embodiment, the cross-layer resource collaborative execution in step 4 is specifically as follows:

[0051] D2D resource pool call: Predicting that UE1 will face high load in time slot 3, activate the adjacent relay UE4 (signal strength > -90dBm) to provide cooperation: ① Physical layer: UE4 enables dual-antenna MIMO mode to share 50% of UE1's uplink traffic; ② MAC layer: Dynamically allocate 20 RB dedicated D2D carriers; ③ Network layer: The core network authorizes UE4 as a temporary gateway for data forwarding.

[0052] Resource release mechanism: In time slot 6, if the actual resource utilization rate is 72% (lower than the reserved threshold of 64%), 16 RBs will be released to the public resource pool. This mechanism avoids long-term resource idleness and improves overall resource utilization.

[0053] (V) Demonstration of the effects of the example

[0054] By monitoring and statistically analyzing data from a complete cycle of this embodiment, the key performance indicators of the "cooperative scheduling scheme" and the "comparison scheme" are compared as follows.

[0055] Table 1. Comparison of Key Performance Indicators

[0056] Performance indicators Comparison with traditional static allocation Cooperative scheduling scheme (this invention) Performance improvement average latency of uRLLC services 28ms 15ms Reduced by approximately 46% eMBB service bandwidth guarantee rate 85% 99.5% An increase of approximately 17% Overall system resource utilization 68% 89% An increase of approximately 31% Business request rejection rate during high load 12% ＜1% Reduced by more than 92% Regional handover service interruption rate 8% 0.5% Reduced by approximately 94%

[0057] The results of this embodiment demonstrate that the network resource adaptive collaborative scheduling communication method provided by this invention effectively solves the problem of resource supply mismatch with actual demand in traditional solutions through dynamic perception and prediction, elastic resource reservation, distributed collaborative scheduling and conflict resolution, and cross-layer resource collaboration. Ultimately, in the complex scenario of a smart park, this invention achieves significant technical effects in ensuring the quality of service for mission-critical communications (uRLLC) (latency reduced by 46%), improving overall resource utilization efficiency (increased by 31%), and significantly enhancing the system's ability to cope with sudden loads (reduced request rejection rate by over 92%), fully demonstrating the superiority and innovation of this invention. The above embodiments verify the effectiveness and advantages of the network resource adaptive collaborative scheduling communication method of this invention in practical applications. In the high-density user scenario of a smart park, this method can accurately predict the changing trends of individual and group resource demands based on the dynamic demand characteristics of different types of UEs and make reasonable resource reservations. Through the dynamic resource scheduling strategy generated by the distributed collaborative scheduling engine, the joint collaborative allocation of base station spectrum resources, edge computing resources, core network bandwidth resources, and adjacent UE relay resources is realized, effectively solving the resource allocation conflict problem.

[0058] During cross-layer resource collaborative execution, the D2D resource pool enables dynamic resource sharing and collaboration between adjacent UEs, improving resource utilization efficiency and reducing system transmission latency and energy consumption. Simultaneously, the resource release mechanism can promptly release reserved resources when actual resource utilization is low, avoiding resource waste and further optimizing resource allocation.

[0059] This invention presents a network resource adaptive collaborative scheduling communication method that excels in ensuring latency stability for high-priority services (such as uRLLC services). It dynamically resolves resource allocation conflicts among multiple nodes through a three-level arbitration mechanism, ensuring the quality of service for latency-sensitive services. In complex scenarios such as smart parks, this method can effectively handle sudden load peaks, reduce resource fragmentation, and provide more stable and efficient communication services for various businesses, demonstrating broad application prospects and promotional value.

[0060] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

[0061] This invention provides a network resource adaptive cooperative scheduling communication method. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. A method of network resource adaptive co-scheduling communication, the method comprising: The method comprises the following steps: Step 1: Establish a dynamic perception layer centered on user equipment (UE), and collect multi-dimensional dynamic demand characteristic data of individual UEs and UE groups in real time; Step 2: Based on historical data and real-time characteristic data, build a demand prediction model to predict the resource demand trend of individual UEs and UE groups in a future time window, and generate a demand prediction graph; Step 3: According to the demand prediction graph, generate a dynamic resource scheduling strategy through a distributed collaborative scheduling engine, which simultaneously acts on the base station RAN, edge computing node MEC, core network control plane function, and adjacent UE equipment; Step 4: Execute the dynamic resource scheduling strategy, and reconfigure resources in multiple layers of the communication protocol stack to complete the joint and collaborative allocation of base station spectrum resources, edge computing resources, core network bandwidth resources, and adjacent UE relay resources.

2. The method of claim 1, wherein, The multi-dimensional dynamic demand characteristics in step 1 include service type characteristics, real-time mobility characteristics, device state characteristics, and channel quality indicators.

3. The method of claim 2, wherein, The collection of dynamic demand characteristics in step 1 includes: obtaining device state characteristics through in-UE sensors, including battery capacity, processor load, and screen activation state; obtaining mobility characteristics through base station measurement reports, including user equipment (UE) motion speed, motion trajectory prediction, and cell handover frequency; identifying service type characteristics through deep packet inspection (DPI), including classification labels for delay-sensitive services, bandwidth-intensive services, and computing-intensive services; and obtaining instantaneous channel quality indicators through channel state information (CSI) feedback.

4. The method of claim 1, wherein, The demand prediction model in step 2 uses a spatio-temporal joint modeling method: for individual UEs, a LSTM-based time series prediction model is used, which inputs historical resource usage sequences and real-time feature vectors, and outputs bandwidth demand, computing resource demand, and maximum tolerated delay for the next N time slots; for UE groups, a graph neural network (GNN) model is used to construct a dynamic topology graph of geographically adjacent UEs, and node feature aggregation is used to predict group resource demand peaks and regional hot spot distribution.

5. The method of claim 4, wherein, The demand prediction model in step 2 includes an elastic resource reservation mechanism. According to the resource demand peaks identified in the demand prediction graph, resources are dynamically reserved at the base station, edge node, and core network in a reservation ratio to cope with predicted resource demand peaks and improve service stability. The calculation formula for the reservation ratio P is: P = k × predicted peak usage rate + (1-k) × historical maximum usage rate, where k is the prediction confidence weight, and the value range is 0.6-0.9; when the actual resource usage rate is lower than 80% of the reservation threshold for n consecutive time slots, the reserved resources are automatically released to the public resource pool.

6. The method of claim 1, wherein, The generation of the dynamic resource scheduling strategy in step 3 contains a joint optimization target, and the objective function is defined as: minimizing the total system overhead = alpha x transmission delay + beta x energy consumption + gamma x resource fragmentation rate; wherein alpha, beta and gamma are weight coefficients, the value range of each is [0, 1], and alpha + beta + gamma = 1, used to balance the relative importance of each optimization target; the constraint conditions include: the number of available RBs of the base station, the computing capacity of the edge node, the maximum bandwidth of the core network, and the upper limit of the relay capability of the adjacent UE; and the optimal resource allocation scheme is solved by an improved distributed Q-learning algorithm.

7. The method of claim 6, wherein, The improved distributed Q-learning algorithm contains a conflict resolution mechanism, used to solve the resource allocation decision conflict of multiple distributed scheduling nodes, and when the resource allocation schemes of multiple scheduling nodes conflict, a three-level arbitration process is performed.

8. The method of claim 7, wherein, The three-level arbitration process includes: Local arbitration: the edge node adjusts resource allocation according to the UE service priority label [eMBB > uRLLC > mMTC], wherein the label eMBB is enhanced mobile broadband, the label uRLLC is ultra-reliable low-latency communication, and the label mMTC is massive machine type communication; Regional arbitration: the core network SDN controller re-plans the cross-domain routing path to preferentially guarantee delay-sensitive services; Global arbitration: the cloud management platform coordinates the resource migration of multiple edge nodes to balance regional load differences; Wherein, the local arbitration is performed first, if the local arbitration cannot solve the conflict, it is upgraded to regional arbitration, and if the regional arbitration still cannot solve the conflict, it is upgraded to global arbitration.

9. The method of claim 1, wherein, The cooperative scheduling of the resources of adjacent UE devices corresponding to the dynamic resource scheduling strategy in step 4 specifically includes: establishing an inter-UE device-to-device (D2D) cooperative resource pool, and including adjacent UEs that meet a signal strength threshold in the resource pool; when it is predicted that a target UE will enter a high load state, selecting a relay UE from the resource pool, and dynamically allocating idle antenna resources, cache resources and computing resources of the relay UE to the target UE; the resource allocation is implemented through a pre-configured D2D link, and the core network authorizes resource use permission through control plane signaling.

10. The method of claim 1, wherein, The multiple layers of the communication protocol stack in step 4 include: a physical layer, a MAC layer and a network layer.

Images