Cloud side information collaboration and content delivery method for air-based high-mobility network

Abstract

The invention relates to the technical field of wireless communication, and discloses an air-based high-mobility network-oriented cloud side information collaboration and content delivery method, which comprises the following steps of: carrying out joint slicing processing on cache resources and communication resources of service nodes on the basis of flight ad hoc network topology and user demand distribution characteristics; content storage and sharing collaboration of multiple service nodes are realized; dynamically updating backbone layer cache resources and communication resources through a backbone layer cache resource scheduling strategy and an inter-slice resource reconfiguration strategy based on cache slice change and communication resource adjustment; on the basis of the topological relation and the service demand change, the delivery strategy of the backbone nodes is optimized, and the communication resources of the service streams in the cache slices are dynamically reconfigured, so that the communication overhead of the backbone nodes in the slices is reduced. According to the invention, on the premise that the content delivery service quality is guaranteed, the system overhead of the air ad hoc network is reduced.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technologies, and particularly to a cloud-edge information collaboration and content delivery method for an air-based highly mobile network. Background Art

[0002] In view of the dynamic topological characteristics and time-varying communication requirements of an air-based multi-service node ad-hoc network, a cache scheduling architecture based on backbone nodes is usually constructed, and the resource scheduling and allocation of backbone nodes are focused on. As an important part of network resource management, caching is particularly important in UAV ad-hoc networks. When multiple service nodes access the backbone UAV ad-hoc network and request content support, in traditional UAV ad-hoc networks, due to limited link and cache resources, the forwarding ability of nodes, cache space, and link bandwidth capacity are all restricted. Therefore, a reasonable caching strategy is crucial for improving network performance.

[0003] In order to meet the quality of service (QoS) requirements between service nodes, it is necessary to efficiently allocate and reuse limited network links and cache resources. By deploying a distributed caching mechanism between backbone nodes and access nodes and combining edge caching technology, the transmission delay in the network can be effectively reduced, the bandwidth occupancy rate can be decreased, and the hit rate of content and the utilization rate of cache resources can be increased, thereby ensuring the stability and reliability of network communication. Especially in latency-sensitive service scenarios, edge caching can pre-store frequently used content on the backbone node closest to the service node, thus greatly reducing the data transmission path and response time.

[0004] Edge caching technology, with its advantages in differentiated service requirements, QoS guarantee, improving system reliability, and optimizing cache resource allocation, has become one of the key technologies to meet service requirements in the UAV ad-hoc network architecture. Through intelligent scheduling and replacement strategies for cached content, the system can dynamically adapt to time-varying communication requirements, reasonably allocate cache resources, maximize the utilization efficiency of overall network resources, and improve the overall service performance of the ad-hoc network.

[0005] However, traditional edge caching technologies face numerous challenges in the drone-based ad-hoc network architecture. First, traditional edge caching mechanisms usually deploy cache servers on ground infrastructure such as base stations. These ground facilities are fixed in location and cannot flexibly adjust the coverage area of data services. This static deployment method is difficult to adapt to the characteristics of mission drones moving frequently in the vast airspace. Second, when the access location or network topology of service clusters changes, traditional edge caching technologies lack dynamic adaptation capabilities, restricting the flexible allocation of network resources and making it difficult to quickly respond to changes in service requirements. In addition, there are different time-scale differences in resource scheduling for traditional edge caching technologies. Periodic reconfiguration may lead to significant interruptions or handover delays in network slices, further affecting the quality and stability of communication. Currently, the research on edge caching technologies mainly includes the following two types: (1) Research on the Joint Optimization Algorithm of Drone Caching Content Delivery and Trajectory (Reference: Luo Yannan. Research on the Joint Optimization Algorithm of Drone Caching Content Delivery and Trajectory [D]. Beijing University of Posts and Telecommunications, 2024): Starting from user access, the author used stochastic geometry methods to analyze the cache hit rate and energy efficiency of two content-centric cooperative caching schemes different from the traditional ones, the access and caching scheduling schemes of macro base station and micro base station cooperation, and micro base station cooperation for specific system parameters. However, this model did not consider the impact of content slicing. The present invention further innovates by introducing a content re-slicing mechanism in cache resource scheduling to reduce the interruption impact on data transmission due to rapid connection switching and improve the flexibility and adaptability of resource allocation.

[0006] (2) QoE-based Drone Network Deployment and Cache Strategy Optimization Method (Reference: Tang Huanbo, Zheng Hongqiang, Shen Qihang, et al. QoE-based Drone Network Deployment and Cache Strategy Optimization Method [J]. Application Research of Computers, 2023, 40(05): 1473 - 1479.): This paper designed a cellular network model using drones supporting caching for assisted communication. This model uses the combination of drone communication and edge caching for traffic offloading, and maximizes the user QoE by jointly optimizing drone deployment, cache placement, and user association, and evaluates it using the mean opinion score. However, this method has not considered making full use of the cache space of adjacent drones, which may lead to unnecessary waste of resources. Summary of the Invention

[0007] To solve the problem that in mission-driven scenarios, due to limited network resources and insufficient cache resources, it is difficult for drone ad-hoc networks to meet the service requirements of drones such as low latency and high bandwidth. To address this problem, the present invention proposes a cloud-edge information collaboration and content delivery method for air-based high-mobility networks, which can improve the content transmission efficiency by optimizing the allocation and scheduling of cache resources.

[0008] The technical solution adopted by the present invention is as follows: A cloud-edge information collaboration and content delivery method for an air-based high-mobility network, comprising: Based on the characteristics of the flying ad-hoc network topology and user demand distribution, jointly slice the cache resources and communication resources of service nodes to achieve content storage and sharing collaboration among multiple service nodes; Based on the changes in cache slicing and communication resource adjustment, through the backbone layer cache resource scheduling strategy and the inter-slice resource reconfiguration strategy, dynamically update the backbone layer cache resources and communication resources; Based on the changes in topological relationships and service requirements, optimize the delivery strategy of backbone nodes, and dynamically reconfigure the communication resources of each service flow within the cache slice, thereby reducing the communication overhead of backbone nodes within the slice.

[0009] Further, the jointly slicing the cache resources and communication resources of service nodes based on the characteristics of the flying ad-hoc network topology and user demand distribution to achieve content storage and sharing collaboration among multiple service nodes includes: Aiming at the characteristics of low update frequency of user content requirements and slow change of the access relationship between users and servers under air-to-ground line-of-sight communication, based on the topological structure of the ad-hoc network and the demand distribution characteristics of service nodes in the air, construct a backbone layer cache policy algorithm based on deep reinforcement learning to optimize the content distribution of cache slices in real time and adapt to time-varying service types and demand fluctuations; At the decision-making moment of each large time scale, based on the inter-slice spectrum resource reconfiguration strategy of the particle swarm optimization algorithm, reserve corresponding spectrum resources for different cache slices to ensure the dynamic schedulability of backbone layer communication resources.

[0010] Further, the constructing the backbone layer cache policy algorithm based on deep reinforcement learning and the inter-slice spectrum resource reconfiguration strategy of the particle swarm optimization algorithm to reserve corresponding spectrum resources for different cache slices includes: Collect the movement trajectories and historical request data of all service nodes and extract features; Extract the features of service nodes based on machine learning methods to predict the time-varying service requirements and access relationships of different service nodes; According to the cache slices corresponding to different contents, based on the future trajectories of service nodes and the probability of content data requirements of different service nodes, obtain the traffic intervals of cache slices.

[0011] Further, the movement trajectories and historical request data of all service nodes include the specific backbone nodes accessed by service nodes, access times, and request content types, and the extracted features include the movement patterns, access frequencies, and request content preferences of service nodes.

[0012] Furthermore, based on the cache slice changes and communication resource adjustment, the backbone layer cache resources and communication resources are dynamically updated through the backbone layer cache resource scheduling strategy and the inter-slice resource reconfiguration strategy, including: At the decision-making moment on a large time scale, the cache slice performs content re-scheduling and constructs a backbone layer bandwidth and power allocation algorithm based on deep deterministic policy gradient; considering co-channel interference, the backbone layer bandwidth and power allocation algorithm optimally allocates the bandwidth and power among backbone nodes; By combining the backbone network topology and the cache resource scheduling strategy, jointly adjust the bandwidth and power configuration of backbone nodes to ensure the timeliness and adaptability of the cache slice content in the future time period, so as to cope with the demand changes and fluctuations of the expected service types.

[0013] Furthermore, the construction of the backbone layer bandwidth and power allocation algorithm based on deep deterministic policy gradient, and the joint adjustment of the bandwidth and power configuration of backbone nodes by combining the backbone network topology and the cache resource scheduling strategy, include: Construct a cache scheduling overhead model between backbone nodes for evaluating transmission efficiency, and the cache scheduling overhead model defines a cache content placement overhead function for describing the system initialization; Using the cache scheduling overhead model, a bandwidth and power allocation algorithm at the decision-making moment on a large time scale is constructed based on the deep deterministic policy gradient algorithm.

[0014] Furthermore, the process of cache content placement refers to the process in which, when the backbone node cache space has no cached content during system initialization, the central node uniformly delivers the cached content required by the backbone nodes to the backbone nodes; the energy consumption during the process of the central node delivering the content to the backbone nodes through the wireless channel is the content placement overhead.

[0015] Furthermore, based on the topological relationship and service demand changes, optimize the delivery strategy of backbone nodes and dynamically reconfigure the communication resources of each service flow in the cache slice, including: Construct a cache energy revenue model, and by maximizing the cache revenue of local caches, use a greedy strategy to generate a cache delivery strategy under the premise of considering the computational overhead; Based on the access relationships between different cache slices and multi-service nodes and the demand characteristics of each service node, optimize the delivery strategy of backbone nodes to maximize the cache energy revenue of each backbone node; When the content request of the service node cannot be satisfied, trigger the in-slice resource reconfiguration strategy to dynamically adjust the transmission power and spectrum bandwidth of each service node in the slice to adapt to the interference and fading under different channel conditions.

[0016] Further, the construction of the cache energy revenue model and the triggering of the in-chip resource reconfiguration strategy when the content requests of service nodes cannot be satisfied include: Construct a data delivery overhead model for evaluating the cache revenue of cached content slices in different backbone nodes; Based on the data delivery overhead model, construct a cache content delivery strategy for a single backbone node; When there is a service flow requested by a service node in the cache slice and it cannot be satisfied, trigger the resource reconfiguration within the slice based on the Actor-Critic algorithm, release the service flow that occupies too many resources and allocate it to the service flow with insufficient resources.

[0017] Further, the data delivery overhead model defines a direct hit overhead, including the energy overhead of directly transmitting content by the backbone node when the content requested by the service node comes from the backbone node directly connected to it.

[0018] The beneficial effects of the present invention are as follows: Aiming at the problems of decreased transmission rate and potential service interruption caused by the movement of service nodes, based on the topological structure of the aerial ad hoc network and the distribution characteristics of service node requirements, the present invention realizes the distribution optimization of cache slices and the pre-allocation of spectrum resources through the cache content scheduling strategy and the inter-chip resource reconfiguration strategy. This method ensures transmission efficiency and service continuity by pre-distributing cached content and reserving spectrum resources. Secondly, based on the distribution of different cached contents, using the cache revenue model and combining with the greedy algorithm, the cached content of backbone nodes is reasonably delivered. When the requests of service nodes cannot be satisfied, trigger the in-chip resource reconfiguration strategy to dynamically adjust the bandwidth and power resource configuration of each backbone node in the chip to improve resource utilization efficiency and reduce content transmission delay.

[0019] The present invention introduces a reconstruction scheme at small time scales and large time scales in cache resource scheduling, and allows the forwarding and delivery of cache resources by borrowing the cache space of adjacent backbone nodes during the cache resource delivery process. By reconstructing the energy overhead function, the present invention optimizes the cache resource scheduling strategy, allowing a part of the cache resources to be pre-cached on backbone drones. During the delivery process, different backbone drones can cooperate and transmit jointly to improve the cache hit rate and delivery efficiency. This strategy aims to achieve an effective balance between the moderate occupation of short-term backbone network resources and the communication requirements of long-term service node task requirements. Description of the Drawings

[0020] Figure 1 is a flowchart of a cloud-edge information collaboration and content delivery method for an air-based highly mobile network according to Embodiment 1 of the present invention.

[0021] Figure 2Schematic diagram of a cloud-edge information collaboration and content delivery method for an air-based high-maneuverability network in Embodiment 2 of the present invention.

[0022] Figure 3 Algorithm flowchart of a cloud-edge information collaboration and content delivery method for an air-based high-maneuverability network in Embodiment 2 of the present invention. Detailed implementation manners

[0023] In order to have a clearer understanding of the technical features, objectives, and effects of the present invention, the detailed implementation manners of the present invention are now described. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0024] To facilitate the understanding of the present invention by those of ordinary skill in the art, the following definitions are first made for the technical terms involved in the present invention: 1. Backbone layer In the backbone layer, M unmanned aircraft with relatively fixed communication topologies form a backbone network, and the backbone unmanned aircraft are represented by the set . The communication between the unmanned aircraft in the backbone layer adopts the OFDM method, and each unmanned aircraft can provide real-time data transmission services to the service nodes accessing it.

[0025] 2. Access layer The access layer consists of N service sub-groups, and each service sub-group contains multiple service nodes. Considering that the unmanned aircraft belonging to the same service sub-group are closely arranged in position and require the same type of content, each service sub-group is abstracted as a virtual service node, represented by the set . In the access layer, the unmanned aircraft performing tasks usually generate various content requirements to ensure the smooth execution of the tasks. Since the tasks performed by the service nodes have strong timeliness, higher requirements are put forward for indicators such as the delay of data delivery.

[0026] 3. Dual time scales In the operation of a distributed edge caching system, the types of content required by service nodes mostly remain fixed over a long period of time, while the changes in the topological access relationships caused by node movement and their corresponding communication resource requirements change relatively quickly. Therefore, the scheduling management process can be divided into two different time scales. In the large time scale, the backbone nodes complete the replacement and scheduling of the cached content stored on them. In the small time scale, the backbone nodes provide the corresponding cached content delivery service to the service nodes that access them and request specific content. The duration of the small time scale is the unit time . The large time scale contains rounds of the small time scale, and also includes a decision-making process for cache resource scheduling. The number of continuous time slots of this decision-making process depends on the cache resource scheduling time slots of the backbone nodes, and is denoted by for the number of continuous time slots of the cache resource scheduling process in the th round of the large time scale. Therefore, the number of continuous time slots in the s th round of the large time scale is .

[0027] 4. Content Sharding By reasonably cutting the content and distributing it for storage in multiple backbone drones, and through the efficient cooperation of multiple backbone drones, it is possible to provide a continuous and efficient content delivery service to service nodes during their movement. The content unit sharding length is defined as . For content , can be divided into shards, and the number of shards corresponding to each type of content is represented by the set . The symbol represents the th shard of the content cached by the backbone drone at the th time slot, where . .

[0028] 5. Cache Revenue The content cache revenue of content sharding is defined as the difference between the energy consumption of the backbone node selecting this content shard for data delivery and the energy consumption of selecting the centralized forwarding strategy for delivering this shard. The cache revenue is related to multiple factors such as the number of service nodes accessed by the backbone node, the number of content types requested, and the positions of different service nodes and backbone nodes.

[0029] 6. Cache Slicing Through the management and configuration of the cache resources and communication resources in the backbone layer, different content shards are distributed and cached to multiple service nodes, and the same type of content may be stored in multiple backbone nodes. The set of these backbone nodes can be logically abstracted as the cache slice of this type of content. It is defined that the The set of backbone nodes for class content sharding is the cache slice , the cache slice obtains spectrum resources through the bandwidth allocation strategy between slices , while the bandwidth and power allocation of the backbone nodes within the slice are regulated by the in-slice resource reconfiguration strategy.

[0030] Embodiment 1 As Figure 1 shown, this embodiment provides a cloud-edge information collaboration and content delivery method for an air-based high-mobility network, including: Based on the flying ad-hoc network topology and user demand distribution characteristics, jointly slice the cache resources and communication resources of service nodes to achieve content storage and sharing collaboration among multiple service nodes; Based on the cache slice changes and communication resource adjustments, dynamically update the backbone layer cache resources and communication resources through the backbone layer cache resource scheduling strategy and the inter-slice resource reconfiguration strategy; Based on the topology relationship and business demand changes, optimize the delivery strategy of backbone nodes, and dynamically reconfigure the communication resources of each service flow within the cache slice, thereby reducing the communication overhead of backbone nodes within the slice.

[0031] It should be noted that the method of this embodiment performs sharding processing on content data by analyzing the topology structure of the air ad-hoc network and user demand distribution characteristics. The sharded data is respectively cached in multiple service nodes to meet the cache capacity limit of a single service node; at the same time, the content data may be distributed and cached in multiple service nodes repeatedly or partially overlapped, thereby improving the efficiency of concurrent transmission of content data. The cache resources and communication resources of service nodes are jointly sliced and managed to ensure content storage and sharing collaboration among multiple service nodes.

[0032] In addition, in the dual time-scale resource slicing and scheduling mechanism included in the method of this embodiment, considering the characteristics that the update frequency of user content requirements is relatively low and the access relationship between users and servers under air-to-ground line-of-sight communication changes relatively slowly, a large time scale is adopted to update the cached content on the server. At the same time, considering the highly dynamic and time-varying characteristics of the wireless spectrum environment, the transmit power and spectrum resources in content sharing and transmission are agilely regulated within a small time scale, thereby improving service efficiency and reducing service overhead.

[0033] Preferably, based on the flying ad-hoc network topology and user demand distribution characteristics, jointly slice the cache resources and communication resources of service nodes to achieve content storage and sharing collaboration among multiple service nodes, including: In view of the characteristics of the low update frequency of user content requirements and the slow change of the access relationship between users and servers in line-of-sight communication, based on the topological structure of the ad hoc network in the air and the distribution characteristics of service node requirements, a backbone layer cache policy algorithm based on deep reinforcement learning is constructed to optimize the content distribution of cache slices in real time to adapt to time-varying service types and demand fluctuations; At each decision-making moment of a large time scale, based on the inter-slice spectrum resource reconfiguration strategy of the particle swarm optimization algorithm, corresponding spectrum resources are reserved for different cache slices to ensure the dynamic schedulability of backbone layer communication resources.

[0034] Preferably, a backbone layer cache policy algorithm based on deep reinforcement learning and an inter-slice spectrum resource reconfiguration strategy based on the particle swarm optimization algorithm are constructed to reserve corresponding spectrum resources for different cache slices, including: Collect the movement trajectories and historical request data of all service nodes and extract features; Extract the features of service nodes based on machine learning methods to predict the time-varying service demands and access relationships of different service nodes; According to the cache slices corresponding to different contents, based on the future trajectories of service nodes and the probability of content data requirements of different service nodes, obtain the traffic interval of the cache slices.

[0035] Preferably, the movement trajectories and historical request data of all service nodes include the specific backbone nodes accessed by the service nodes, the access time, and the types of requested contents, and the extracted features include the movement patterns, access frequencies, and content preferences of the service nodes.

[0036] Preferably, based on the changes in cache slices and the adjustment of communication resources, through the backbone layer cache resource scheduling strategy and the inter-slice resource reconfiguration strategy, the backbone layer cache resources and communication resources are dynamically updated, including: At the decision-making moment of a large time scale, the cache slices perform content rescheduling, and a backbone layer bandwidth and power allocation algorithm based on deep deterministic policy gradients is constructed; the backbone layer bandwidth and power allocation algorithm optimally allocates the bandwidth and power between backbone nodes considering co-channel interference; By combining the backbone network topology and the cache resource scheduling strategy, jointly adjust the bandwidth and power configurations of backbone nodes to ensure the timeliness and adaptability of the cache slice content in the future time period, so as to cope with the demand changes and fluctuations of expected service types.

[0037] Preferably, a backbone layer bandwidth and power allocation algorithm based on deep deterministic policy gradients is constructed, and by combining the backbone network topology and the cache resource scheduling strategy, jointly adjust the bandwidth and power configurations of backbone nodes, including: Build a cache scheduling overhead model between backbone nodes for evaluating transmission efficiency. The cache scheduling overhead model defines a cache content placement overhead function for describing the system initialization; Utilize the cache scheduling overhead model to construct a bandwidth-power allocation algorithm at large time-scale decision moments based on the deep deterministic policy gradient algorithm.

[0038] Preferably, the process of placing cache content refers to the process in which, when the cache space of backbone nodes is not placed with cache content during system initialization, the central node uniformly delivers the cache content required by the backbone nodes to the backbone nodes; the energy consumption during the process of the central node delivering content to the backbone nodes through the wireless channel is the content placement overhead.

[0039] Preferably, based on the topological relationship and the change of service requirements, optimize the delivery strategy of backbone nodes, and dynamically reconfigure the communication resources of each service flow in the cache slice, including: Build a cache energy revenue model, and generate a cache delivery strategy using a greedy strategy under the premise of considering the computational overhead by maximizing the cache revenue of local caches; Based on the access relationship between different cache slices and multi-service nodes and the demand characteristics of each service node, optimize the delivery strategy of backbone nodes to maximize the cache energy revenue of each backbone node; When the content request of the service node cannot be satisfied, trigger the in-slice resource reconfiguration strategy, and dynamically adjust the transmission power and spectrum bandwidth of each service node in the slice to adapt to interference and fading under different channel conditions.

[0040] Preferably, building a cache energy revenue model, and when the content request of the service node cannot be satisfied, triggering the in-slice resource reconfiguration strategy, including: Build a data delivery overhead model for evaluating the cache revenue of cache content sharding in different backbone nodes; Based on the data delivery overhead model, construct the cache content delivery strategy of a single backbone node; When there is a service flow requested by a service node in the cache slice and it cannot be satisfied, trigger the in-slice resource reconfiguration based on the Actor-Critic algorithm, release the service flow that occupies too many resources and allocate it to the service flow with insufficient resources.

[0041] Preferably, the data delivery overhead model defines a direct hit overhead, including the energy overhead of directly transmitting content by the backbone node when the content requested by the service node comes from the backbone node directly connected to it.

[0042] Specifically, the method of this embodiment can be implemented by the following steps: A1. Slice the content data by analyzing the topology of the ad-hoc network in the air and the distribution characteristics of user requirements, and dynamically optimize the cache content policy based on energy consumption and the inter-slice resource reconfiguration policy to respond to the dynamic changes in the access relationships and service requests of time-varying service requirements. Jointly slice and manage the cache resources and communication resources of the service nodes to improve the concurrent delivery efficiency of the content data and ensure the content storage and sharing collaboration among multiple service nodes; A2. In each large time scale, cache resource delivery is carried out by embedding sub-periods of multiple small time scales to achieve a rapid response to immediate service requirements. The duration of the large time scale is set to be non-fixed and can be flexibly adjusted according to the dynamic changes in the cache content scheduling policy; A3. In the process of cache resource scheduling at the backbone layer, based on the characteristics of future cache content policy changes, by configuring the bandwidth and power of the backbone nodes, the backbone layer can obtain a better cache content distribution state with less energy consumption. That is, by allocating the bandwidth and power of the backbone nodes for cache content scheduling, the flexibility of cache content slicing is ensured, and the energy consumption of cache content scheduling is reduced; A4. In the process of inter-slice resource reconfiguration of the backbone layer cache slices, based on the dynamic changes in the future network access topology relationship and cache content, spectrum resources are pre-allocated to each cache slice to better provide data support services for the service layer. At the same time, considering the characteristics of the low update frequency of user content requirements and the slow change of the user-server access relationship in line-of-sight communication in the air, a large time scale is adopted to update the inter-slice resource reconfiguration.

[0043] A5. Based on the comprehensive evaluation of service performance such as throughput and transmission energy consumption, the sending list of cache slices in the small time scale is dynamically updated. In the small time scale, by selecting the delivery strategies of cache content of different backbone nodes to make it as suitable as possible for service requirements, the purpose of reducing energy consumption and service delay requirements is achieved.

[0044] A6. In view of the highly time-varying and dynamic characteristics of the wireless spectrum environment, within a small time scale, agilely adjust the transmission power and spectrum resource allocation in the process of content sharing and transmission to cope with the rapid changes in the spectrum environment, ensure the stability of the communication link and the quality of service. Specifically, by real-time monitoring the in-slice spectrum usage situation, dynamically adjust the transmission power and spectrum bandwidth of each service node so that it can adapt to the interference and fading under different channel conditions, while reducing the waste of spectrum resources, improving the overall transmission efficiency, reducing the transmission energy consumption and ensuring the timeliness and reliability of data transmission.

[0045] In summary, the method of this embodiment reduces the system overhead of the ad-hoc network in the air on the premise of ensuring the quality of content delivery service. Different from the existing edge caching methods, in the method of this embodiment, based on the ad-hoc network topology in the air and the characteristics of user demand distribution, content data may either be cut and cached separately in multiple service nodes to meet the cache space constraints of a single service node, or be distributed in multiple service nodes with repetition or partial overlap, thereby improving the concurrent delivery efficiency of content data.

[0046] In the process of multi-point collaborative edge caching and content delivery for the ad-hoc network in the air, joint slicing processing is performed on the cache resources and communication resources of service nodes, ensuring content storage and sharing collaboration among multiple service nodes.

[0047] In resource slicing resource scheduling, considering the characteristics that the update frequency of user content requirements is relatively low and the user-server access relationship in line-of-sight communication in the air changes relatively slowly, the server cache content is updated at a large time scale; considering the highly time-varying and dynamic characteristics of the wireless spectrum environment, the transmit power and spectrum resources in content sharing and delivery are adjusted agilely at a small time scale to improve service efficiency and reduce service overhead.

[0048] Embodiment 2 This embodiment is based on Embodiment 1: As Figure 2 shown, this embodiment provides a cloud-edge information collaboration and content delivery method for an air-based highly mobile network, including the following steps: S1. Considering the characteristics that the update frequency of user content requirements is relatively low and the user-server access relationship in line-of-sight communication in the air changes relatively slowly, based on the topological structure of the ad-hoc network in the air and the distribution characteristics of service node requirements, a backbone layer cache policy algorithm based on deep reinforcement learning (DQN) is constructed to optimize the content distribution of cache slices in real time and adapt to time-varying service types and demand fluctuations. At each decision moment of a large time scale, an inter-slice spectrum resource reconfiguration strategy based on the particle swarm optimization algorithm is adopted to reserve appropriate spectrum resources for different cache slices, ensuring the dynamic schedulability of backbone layer communication resources, so as to quickly respond to changes in service requirements.

[0049] S2. At the decision moment of a large time scale, the cache slices perform content re-scheduling, and a backbone layer bandwidth and power allocation algorithm based on deep deterministic policy gradient (DDPG) is constructed. This algorithm optimally allocates the bandwidth and power between backbone nodes considering co-channel interference to achieve lower energy consumption during the process of optimizing the cache content distribution. By combining the backbone network topology and the cache resource scheduling strategy, the bandwidth and power configurations of backbone nodes are jointly adjusted to ensure the timeliness and adaptability of the cache slice content in the future time period, so as to cope with the demand changes and fluctuations of expected service types.

[0050] S3. Build a cache energy gain model, aiming to generate an efficient cache delivery strategy in a short time by maximizing the cache gain of local caching and using a greedy strategy while considering the computational overhead. Based on the access relationships between different cache slices and multi-service nodes and the demand characteristics of each service node, optimize the delivery strategy of backbone nodes to maximize the cache energy gain of each backbone node and ensure the timeliness of cached content. When the QoS of service nodes cannot be guaranteed, trigger an in-chip resource reconfiguration strategy based on the Actor-Critic algorithm, dynamically adjust the transmission power and spectrum bandwidth of each service node in the chip to adapt to interference and fading under different channel conditions, reduce spectrum resource waste, improve the overall transmission efficiency, reduce transmission energy consumption, and ensure the timeliness and reliability of data transmission.

[0051] In step S1, build a cache strategy model at the large time-scale decision moment, dynamically update the cache content distribution of backbone nodes, and design an inter-chip resource reconfiguration strategy based on the particle swarm optimization algorithm to reserve an appropriate amount of spectrum resources for different cache slices. Preferably, it specifically includes the following sub-steps: S 11. Collect the movement trajectories and historical request data of all service nodes, specifically including the specific backbone nodes accessed by service nodes, access times, and request content types, and extract features from these data, including the movement patterns, access frequencies, and request content preferences of service nodes.

[0052] S 12. Use a machine learning method based on Double-DQN to extract the features of service nodes to predict the time-varying service demands and access relationships of different service nodes. More preferably, step S 12 specifically includes the following sub-steps: S 121. The machine learning algorithm is restricted to obtain the state of the current environment. The state at the k th large time-scale decision moment is defined as , where the state of service node consists of the position T of the th small time-scale round and the request probability distribution model , where represents the three-dimensional spatial coordinates of service node at the k th time slot. During the execution of tasks, service nodes will generate corresponding content data demands. Assume that the probability that service node generates a demand for content k at the th time slot is . Based on service nodes Position Based on the positions of each backbone node and the downlink bandwidth power, the service node at the k future access relationship table for the small time scale can be obtained. These features can be comprehensively defined as ; for the backbone node , its state is composed of its own position , topological relationship and cache matrix . These features can be represented by .

[0053] S 122. Based on the current state space, the machine learning algorithm outputs a decision , where is the set of cache space variables for each backbone node , reflecting the distribution of the cached content in the backbone layer.

[0054] S 123. Execute the decision obtained by the machine learning algorithm to obtain the reward given by the current environment. The reward function for the th large time scale is specifically defined as , where is the expected transmission energy overhead of the system at the th large time scale, that is, the expected energy transmission overhead of the system calculated by the centralized transmission strategy at the th large time scale. is the sum of the link overhead of the backbone layer and the transmission overhead of the central node at the kth large time scale, that is, the content replacement overhead; is the cache content delivery overhead from the backbone node to the service node at the kth large time scale. The specific physical meaning of the reward function is: when the service node requests content, the difference in energy overhead between the data delivery strategy of directly transmitting by the central server and the edge caching strategy, that is, the energy overhead that can be saved by adopting the current strategy compared to the centralized strategy.

[0055] S 124. When the environment enters the next state, store the state space, action, reward of the previous moment and the state of the current moment in the experience pool, and take out a certain number of samples from the experience pool to update the behavior network , and the loss function used in its update process is as shown in Equation (1): (1) where is specifically defined as shown in Equation (2): (2) In the formula, represents the value prediction of the action taken by the current network in state ; is the target value for learning and updating; is the reward obtained at time step t, is the discount factor, is the next state reached after taking the action ; is the action selected by the target network; is the parameter of the current network, is the parameter of the target network.

[0056] S 125. When the caching policy is executed a certain number of times, the parameters of the behavior network are assigned to the target network .

[0057] S 13. By following step S 12, caching slices corresponding to different contents are obtained. Based on the future trajectories of service nodes and the content data requirements of different service nodes, probabilities are generated, and thus the caching slice can be obtained, that is, the set of backbone nodes caching the th type of content shard, and the traffic interval of this set is . More preferably, the following steps are carried out according to the traffic intervals of different caching slices: S 131. Construct an optimization model for the inter-slice resource reconfiguration policy based on the particle swarm optimization algorithm, and the specific definition is as shown in Equation (3): (3) where is the network benefit generated by unit satisfaction; is the bandwidth resource of the caching slice before reconfiguration; is the bandwidth resource reallocated by the backbone layer to the caching slice ; the unit reconfiguration cost of the caching slice in the backbone layer is ; is an auxiliary variable, and is equal in dimension to .

[0058] S ​132. Initialize the population in the particle swarm algorithm according to the optimized model in step S 131. Randomly generate particles in the solution space of the problem, and each particle represents a candidate solution. The positions and velocities of the particles are randomly initialized, and at the same time, it is judged whether the particle positions satisfy the constraint conditions. If not, they are regenerated to ensure that all particles in the initial population satisfy the constraint conditions. Set the size of the particle swarm to , the maximum number of iterations , and parameters such as the learning factors and etc.

[0059] S 133. Construct the fitness function as follows: (4) where is the initial objective function of the constrained optimization problem, is the penalty factor at the th iteration, and is the penalty term.

[0060] S 134. Based on the fitness function constructed in step S 133, calculate the fitness of each particle. For each particle, compare its current fitness value with the best fitness value in the history of the particle. If the current fitness value is better, update the individual best position of the particle , and find the current global optimal solution among all particles and use it as the current global optimal position .

[0061] S 135. Update the velocities and positions of the particles (5) (6) Equation (5) is the velocity update formula, and equation (6) is the position update formula. is the number of the th particle in the particle swarm, k represents the number of iterations, is the inertia weight, and its value is a non - negative real number, reflecting the degree to which the next velocity is affected by the current velocity. Generally, take , and at this time, the performance of the particle swarm algorithm is the best. and are the learning factors, is the individual learning factor, indicating the experience and thinking of the particle from its own historical data; is the social learning factor of each particle, indicating the cooperation and information sharing among particles. and is a random number. represents the optimal position of the i th particle during multiple updates. It can be seen that represents the direction in which the particle points from its current position to its historical optimal position. is the current global optimal position. It can be seen that represents the direction in which the particle points from its current position to the current global optimal position. The particle continuously updates the individual extreme value and the global optimal value within the feasible region until the termination condition is reached.

[0062] S 136. Determine whether the termination condition is reached, such as whether the maximum number of iterations is reached , or whether the change in the global optimal solution is less than the preset threshold. If the termination condition is reached, output the global optimal solution; otherwise, return to step S 134 to continue the iteration. In this project, the number of iterations is used for judgment, that is, when the particle swarm algorithm iterates to the algorithm is terminated.

[0063] Step S 2. Based on the limited bandwidth and power resources of the backbone nodes, construct an overhead model for energy transmission between backbone nodes, and based on this model, construct a bandwidth-power allocation algorithm for cache scheduling of backbone nodes. Preferably, it specifically includes the following sub-steps: S 21. Construct a cache scheduling overhead model between backbone nodes for evaluating transmission efficiency.

[0064] S 211. This model defines a cache content placement overhead function for describing the system initialization. The cache content placement process refers to the process in which the central node uniformly delivers the cache content required by the backbone nodes to the backbone nodes because the cache space of the backbone nodes is not placed with cache content during system initialization. The energy consumption during the process of the central node delivering the content to the backbone nodes through the wireless channel is the content placement overhead : (7) Among them, is the content fragment size, represents the backbone node requesting cache fragments at system initialization.

[0065] S212. During the continuous update of the caching policy, the system needs to achieve the caching scheduling policy with lower energy consumption based on the caching content distribution status of the current backbone nodes through link allocation and caching content scheduling among the backbone nodes. In this process, each backbone node can request the required caching content from adjacent backbone nodes or from the central node. The energy consumption of this process is specifically expressed as Equation (8): (8) where is the set of backbone nodes, is the set of caching content, is the caching content corresponding number of shards; represents the backbone node j at the th time slot, the th shard of the caching content indicator variable; is the transmission power of the backbone node at the th time slot; is the connection indicator variable between backbone nodes, indicating that a communication link has been established between the backbone node and ; is the content shard size; is the energy consumption of the central server for transmitting a unit data volume.

[0066] S 22. Using the energy consumption model constructed in Step S 21, a bandwidth power allocation algorithm at the large time-scale decision moment is constructed based on the DDPG algorithm. More preferably, it specifically includes the following sub-steps: S 221. Take the action space S defined in Step 122 as the state space at the current moment and input it into the bandwidth power allocation algorithm based on the DDPG algorithm.

[0067] S 222. The bandwidth power allocation algorithm obtains the current action based on the state at the current moment, where represents the action space of the backbone node at the k th large time-scale, represents the transmission power of the backbone node at the k th large time-scale, represents the backbone node Channels available in this large time scale round.

[0068] S 223. After that, the environment gives the reward under the current decision to guide the parameter update in the bandwidth power algorithm. Among them, the reward is calculated through the reward function . That is the cache content replacement cost, which can be specifically obtained through Equation (8); Indicates the energy cost obtained by averaging through multiple solutions using the random policy, so as to ensure that the system moves in the direction of reducing the energy cost.

[0069] S 224. The system enters the next state. At this time, the state at the previous moment, the executed action, the reward, and the state at the current moment are stored in the experience pool for updating the neural network parameters. When the bandwidth power allocation algorithm is executed a certain number of times, the system takes out a certain number of samples from the experience pool and uses the mean squared error loss function , with the learning rate to update Actor all the parameters of the current network , and uses the loss function , with the learning rate to update Critic all the parameters of the current network .

[0070] Among them, is the loss function of the Actor; is the action selection given by the Actor network according to its policy function in the state ; is the value estimated by the Critic network in the state after executing the action , representing the expected value of the future return; is the loss function of the Critic; is the target value.

[0071] S 225. When Actor the current network and Critic the current network are updated a certain number of times, update Actor and Critic the target network parameters: .

[0072] Step S3 Build a data delivery energy overhead model at a small time scale, calculate the cache content benefit based on this model, and then design the transmission strategy of the backbone node, aiming to improve the cache hit rate and reduce the energy consumption of data delivery. When the content request of the service node cannot be satisfied, trigger the on-chip resource reconfiguration strategy to quickly adjust the transmission power and spectrum resources of content sharing and transmission within a small time scale, so as to improve the service efficiency and reduce the service cost. Preferably, it specifically includes the following steps: S 31. Build a data delivery overhead model for evaluating the cache benefit of cached content slices at different backbone nodes. More preferably, it specifically includes the following sub-steps: S 311. This model defines a direct hit overhead, which is specifically manifested as the energy overhead of directly transmitting content by the backbone node when the content requested by the service node comes from the backbone node directly connected to it. The transmission overhead of its k th small time scale is calculated as shown in Equation (9): (9) where, represents the type of content selected for transmission by the backbone node at the k th small time scale, that is indicates that the backbone node selects to transmit content at the th slice; is the transmission power allocated by the system to the backbone node ; represents the cache of the backbone node at the k th small time scale ; is the communication rate between the backbone node and the service node k at the th small time scale, which is specifically calculated by Equation (10): (10) where, is the unit channel bandwidth of the backbone layer, is the set of communication channels in the backbone layer. represents that the k th backbone layer channel is allocated to the backbone node at the th time slot. is the transmission power of the backbone node at the th time slot, Backbone Node j The additive Gaussian noise power at the location, Backbone Node With backbone nodes j The path loss between , Backbone Node With backbone nodes j In addition, .

[0073] S 312, No. k Indirect forwarding overhead on a time scale of hours The calculation process is shown in formula (11): (11) in, , For backbone nodes A set of directly connected backbone nodes. , reflecting the backbone nodes Select to forward the content after requesting it from adjacent nodes.

[0074] S 313. When the cached content does not hit, the central server directly transmits the content, and its energy consumption As shown in formula (12): (12) Among them, the constraints Mark the first small time scale starting moment at the beginning of each large time scale. is a positive integer; Indicates a business node s The indicator function that the request content reaches the maximum delay tolerance.

[0075] S 314. Based on steps S311 to S313, The cache content delivery overhead of time slots can be expressed as: (13) in, is the direct delivery overhead, which can be calculated by formula (9); is the indirect forwarding overhead, which can be calculated by formula (11); is the forwarding overhead of the central server, which can be obtained by formula (12).

[0076] S 32. Based on SThe data delivery overhead model constructed in 31 is used to construct the cache content delivery strategy for a single backbone node. For the backbone node in terms of k the th small time scale, the backbone node selects the energy overhead of delivering the cache content slice , more preferably, the specific calculation process is shown in Equation (14): (14) In Equation (14), , is the set of adjacent backbone nodes of the backbone node ; , is the k th small time scale, the set of service nodes of the service nodes that access the backbone node and request the th shard of the s type of content. In this formula, indicates that the backbone node delivers the cache content by direct broadcast, and represents that the backbone node delivers the cache content by forwarding the broadcast.

[0077] Based on Equation (14), the local optimization objective of the greedy strategy can be constructed, as shown in Equation (15).

[0078] (15) Among them, represents the shard content selected by the backbone node to send at the k th small time scale; is the set of sending strategies for each backbone node . C1 indicates the numerical constraint; C2 indicates that this optimization objective is carried out on a small time scale. Equation (15) indicates that the strategy needs to solve the maximum cache benefit of each round of small time scales, and the sending strategy of each round of backbone nodes can be obtained by maximizing the local cache benefit of a single backbone node.

[0079] S 33. When there is a service flow requested by a service node in the cache slice and it cannot be satisfied, the resource reconfiguration within the slice based on Actor-Critic is triggered, releasing the service flow that occupies too much resources and allocating it to the service flow with insufficient resources, which specifically includes the following sub-steps: S 331. The Actor-Critic algorithm needs to obtain the current environmental state. Use the symbol to represent the th time slot allocated by the system to the cache slice The bandwidth of represents the th time slot accessing the cache slice service node of , The cache slice The required data volume of, denoted by the symbol represents the th time slot cache slice backbone node of , The cache slice The data volume. In summary, the state space can be represented by the triple , where , , .

[0080] S 332. Based on the state space defined in step S 331, input it into the trained policy to select an action , and determine whether the action satisfies the constraint conditions and . If not, regenerate the action . Among them, and are the bandwidth and power of the backbone UAV in the cache slice communicating with other backbone nodes in the k th time slot respectively. And the sum of the bandwidth resources of each backbone node cannot exceed the resources allocated by the system to the cache slice, that is , and the power of each backbone node in each slice cannot exceed its maximum power constraint, that is , .

[0081] S 333. Obtain the reward value and the next moment state from the current environment. The reward value can be obtained from the of the cache slice . Among them, represents the th time slot cache slice network benefit, is the th time slot service node in the cache slice The satisfaction with the services provided and each being the nth time-slot backbone node The magnitude of the change in bandwidth and power before and after within the cache slice The final reward value can be expressed by Equation as follows

[0082] Finally, based on the above steps, an overall algorithm model can be obtained, as shown in Figure 3 the following

[0083] It can be seen from the above embodiments that the method of the present invention improves the communication performance of the UAV ad-hoc network. Through this method, the cache hit rate is effectively increased, the communication delay is reduced, and the timeliness of the service nodes to obtain the data required for tasks is ensured

[0084] Embodiment 3 Based on Embodiment 1, this embodiment provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the method for cloud-edge information collaboration and content delivery for an air-based highly mobile network in Embodiment 1. Among them, the computer program can be in the form of source code, object code, executable file or some intermediate form, etc

[0085] Embodiment 4 Based on Embodiment 1, this embodiment provides a computer-readable storage medium, storing a computer program, and when the computer program is executed by a processor, it implements the method for cloud-edge information collaboration and content delivery for an air-based highly mobile network in Embodiment 1. Among them, the computer program can be in the form of source code, object code, executable file or some intermediate form, etc. The storage medium includes: any entity or device capable of carrying the computer program code, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the storage medium does not include electrical carrier signals and telecommunication signals

[0086] It should be noted that, for the foregoing method embodiments, for the sake of simplicity of description, they are expressed as a series of action combinations. However, those skilled in the art should be aware that this application is not limited by the described action sequence, because according to this application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

Claims

1. A cloud-edge information collaboration and content delivery method for an air-based highly mobile network, characterized in that Including: Based on the characteristics of the flying ad-hoc network topology and user demand distribution, jointly slice the caching resources and communication resources of service nodes to achieve content storage and sharing collaboration among multiple service nodes; Based on the changes in caching slices and communication resource adjustment, dynamically update the backbone layer caching resources and communication resources through the backbone layer caching resource scheduling strategy and the inter-slice resource reconfiguration strategy; Based on the changes in topological relationships and service requirements, optimize the delivery strategy of backbone nodes and dynamically reconfigure the communication resources of each service flow within the caching slice, thereby reducing the communication overhead of backbone nodes within the slice.

2. The cloud-edge information collaboration and content delivery method for an air-based highly mobile network according to claim 1, wherein The step of jointly slicing the caching resources and communication resources of service nodes based on the characteristics of the flying ad-hoc network topology and user demand distribution to achieve content storage and sharing collaboration among multiple service nodes includes: Aiming at the characteristics of low update frequency of user content requirements and slow change of the access relationship between users and servers under line-of-sight communication, based on the topological structure of the aerial ad-hoc network and the distribution characteristics of service node requirements, construct a backbone layer caching strategy algorithm based on deep reinforcement learning to optimize the content distribution of caching slices in real time and adapt to time-varying service types and demand fluctuations; At the decision-making moment of each large time scale, based on the inter-slice spectrum resource reconfiguration strategy of the particle swarm optimization algorithm, reserve corresponding spectrum resources for different caching slices to ensure the dynamic schedulability of backbone layer communication resources.

3. A cloud-edge information collaboration and content delivery method for an air-based highly mobile network according to claim 2, characterized in that, The step of constructing a backbone layer caching strategy algorithm based on deep reinforcement learning and an inter-slice spectrum resource reconfiguration strategy of the particle swarm optimization algorithm to reserve corresponding spectrum resources for different caching slices includes: Collect the movement trajectories and historical request data of all service nodes and extract features; Extract the features of service nodes based on machine learning methods to predict the time-varying service requirements and access relationships of different service nodes; Based on the caching slices corresponding to different contents, obtain the traffic interval of the caching slice based on the future trajectories of service nodes and the probability of content data requirements of different service nodes.

4. A cloud-edge information collaboration and content delivery method for an air-based highly mobile network according to claim 3, characterized in that The movement trajectories and historical request data of all service nodes include the specific backbone nodes accessed by the service nodes, the access time, and the types of requested contents, and the extracted features include the movement patterns, access frequencies, and content request preferences of service nodes.

5. A cloud-edge information collaboration and content delivery method for an air-based high-maneuver network according to claim 1, characterized in that The step of dynamically updating the backbone layer caching resources and communication resources based on the changes in caching slices and communication resource adjustment through the backbone layer caching resource scheduling strategy and the inter-slice resource reconfiguration strategy includes: At the decision-making moment of the large time scale, the caching slice performs content rescheduling and constructs a backbone layer bandwidth and power allocation algorithm based on deep deterministic policy gradients; the backbone layer bandwidth and power allocation algorithm optimally allocates the bandwidth and power between backbone nodes considering co-channel interference; By combining the backbone network topology and the caching resource scheduling strategy, jointly adjust the bandwidth and power configurations of backbone nodes to ensure the timeliness and adaptability of the caching slice content in the future time period, thereby coping with the demand changes and fluctuations of the expected service types.

6. The cloud-edge information collaboration and content delivery method for an air-based highly mobile network according to claim 5, wherein Construct the backbone layer bandwidth and power allocation algorithm based on deep deterministic policy gradient, and jointly adjust the bandwidth and power configuration of backbone nodes by combining the backbone network topology and cache resource scheduling strategy, including: Construct a cache scheduling overhead model between backbone nodes for evaluating transmission efficiency, and the cache scheduling overhead model defines a cache content placement overhead function for describing the system initialization. Utilize the cache scheduling overhead model to construct a bandwidth and power allocation algorithm at large time-scale decision moments based on the deep deterministic policy gradient algorithm.

7. A cloud-edge information collaboration and content delivery method for an air-based high-mobility network according to claim 6, characterized in that The process of cache content placement refers to the process in which the central node uniformly delivers the cache content required by the backbone nodes to the backbone nodes when the cache space of the backbone nodes is not placed with cache content during system initialization; the energy consumption during the process of the central node delivering the content to the backbone nodes through the wireless channel is the content placement overhead.

8. A cloud-edge information collaboration and content delivery method for an air-based high-mobility network according to claim 1, characterized in that Optimize the delivery strategy of backbone nodes based on the changes in topological relationships and service requirements, and dynamically reconfigure the communication resources of each service flow within the cache slice, including: Construct a cache energy revenue model, and generate a cache delivery strategy using a greedy strategy while considering the computational overhead by maximizing the cache revenue of local caches. Optimize the delivery strategy of backbone nodes based on the access relationships between different cache slices and multi-service nodes and the demand characteristics of each service node to maximize the cache energy revenue of each backbone node. When the content request of a service node cannot be satisfied, trigger the in-slice resource reconfiguration strategy to dynamically adjust the transmission power and spectrum bandwidth of each service node within the slice to adapt to interference and fading under different channel conditions.

9. A cloud-edge information collaboration and content delivery method for an air-based highly mobile network according to claim 8, characterized in that The construction of the cache energy revenue model and the triggering of the in-slice resource reconfiguration strategy when the content request of a service node cannot be satisfied include: Construct a data delivery overhead model for evaluating the cache revenue of cache content slices in different backbone nodes. Based on the data delivery overhead model, construct a cache content delivery strategy for a single backbone node. When there is a service flow requested by a service node in the cache slice and it cannot be satisfied, trigger the in-slice resource reconfiguration based on the Actor-Critic algorithm, release the service flow that occupies too many resources and allocate it to the service flow with insufficient resources.

10. A cloud-edge information collaboration and content delivery method for an air-based highly mobile network according to claim 9, characterized in that, The data delivery overhead model defines a direct hit overhead, including the energy overhead of directly transmitting the content by the backbone node when the content requested by the service node comes from the backbone node directly connected to it.

Images