A service-aware resource coordination allocation method

Abstract

The application claims a service-aware resource collaborative allocation method, belonging to the field of mobile edge computing resource allocation, which comprises: analyzing task demand characteristics, obtaining task abstract categories combined with task demand characteristics, and mapping them into corresponding demand types. Based on the task demand type, candidate collaborative node clusters are selected through destruction operators and repair operators, then the task and resource matching problem is modeled as a combinatorial optimization problem, the collaborative benefits of candidate node combinations are calculated, and the edge node combination with the highest collaborative benefit is selected. The application mainly considers the problems of task demand fuzziness and task and resource matching imbalance in the computing power network, designs a task demand type identification algorithm and a resource collaborative allocation strategy, selects a reasonable collaborative node cluster, and reduces the time delay while improving the resource utilization.

Description

Technical Field

[0001] This invention relates to the field of resource allocation in mobile edge computing, and particularly to a method for task requirement identification and collaborative resource allocation. Background Technology

[0002] The development of edge intelligence, the Internet of Things (IoT), and computing networks has driven the rapid growth of the digital economy. With the massive increase in the number of application scenarios and tasks, users' demands for latency and other service quality indicators are gradually increasing. Effectively sensing task requirements at the edge server level and identifying fuzzy task requirements as corresponding task types presents a challenge. Simultaneously, achieving collaborative resource allocation among edge servers is also a problem that requires current attention.

[0003] In current research, researchers typically assume that the type of service request is known. However, in reality, service request types are often not specific enough, preventing servers from directly obtaining the accurate type of service request, thus leading to a decrease in service quality. For example, edge servers can only monitor basic information about task resources but do not explicitly state whether the task requires real-time or non-real-time services, resulting in an unclear service request type. Therefore, edge servers cannot directly determine the user's current service request type, leading to reduced service quality and a negative impact on user experience.

[0004] In resource allocation research, hierarchical optimization algorithms, such as distributed deep reinforcement learning and hierarchical genetic algorithms, are often used to achieve resource allocation. These are typically applied in smart cities and multi-layered architectures involving cloud, edge nodes, and sensing devices. However, resource allocation suffers from an imbalance between task and resource matching. The lack of a reasonable collaborative node cluster leads to optimization problems in matching task requirements with node resources. This results in some nodes becoming overloaded and inefficient, while other nodes waste resources. This imbalance leads to reduced resource allocation efficiency, increased latency, and decreased resource utilization. In conclusion, with the growing awareness of improving service quality and user experience, the importance of task requirement perception and collaborative resource allocation is increasingly recognized. It is precisely because of this research background and these challenges that research on task requirement perception and collaborative resource allocation becomes particularly important.

[0005] Therefore, this invention proposes a service-aware resource collaborative allocation method.

[0006] A search revealed that application publication number CN120456123A belongs to the field of mobile communication technology resource allocation. The method includes providing a task offloading and resource allocation method for edge computing, with the following steps: Constructing a three-layer network model: Constructing a cloud-edge-device three-layer architecture, including M terminals, N edge servers, and 1 cloud server, specifying wireless transmission from the terminal to the edge and wired transmission from the edge to the cloud; Defining key model parameters: Determining terminal task load, offloading ratio, and CPU usage. Considering terminal mobility and dynamic task generation, a quintuple is used to describe the terminal offloading strategy. Offloading and priority rules are determined using a partial offloading method, dynamically adjusting task priorities based on remaining edge server resources, and then offloading according to the new priority order in a first-in-first-out (FIFO) manner. Computation and communication models are established, constructing local, edge, and cloud computing models, as well as terminal-edge and edge-cloud communication models, calculating latency and energy consumption at each stage. An optimization objective function is defined, aiming to minimize the weighted sum of total task latency and total energy consumption. Latency and energy consumption are balanced using weight coefficients to construct the system's total cost function. Optimization constraints are added, limiting the offloading ratio range, ensuring total latency does not exceed tolerance values, total energy consumption does not exceed thresholds, and ensuring that terminal, edge, and cloud computing power do not exceed limits. The problem is transformed into an MDP, defining a quintuple of states and actions, and employing the MAPPO algorithm with an Actor-Critic dual network. The optimal strategy is output through MAPPO algorithm optimization convergence, outputting the optimal offloading ratio, resource allocation, and bandwidth allocation scheme to minimize system cost.

[0007] The difference between this invention and the methods described above lies in the fact that it considers collaborative resource allocation within an edge server node collaborative cluster, improving allocation efficiency through resource complementarity and dynamic scheduling among multiple nodes in the cluster. For mobile edge scenarios, the aforementioned methods, which rely on independent resource allocation from a single edge server or the cloud, may suffer from problems such as insufficient single-edge coverage due to terminal mobility, limited single-node resources leading to overload, and a lack of inter-node resource complementarity scheduling. This is not universally applicable to latency-sensitive and computationally resource-constrained mobile edge scenarios. Furthermore, to address the need to build edge server node collaborative clusters in edge computing scenarios to improve resource utilization and service reliability, the design of collaborative mechanisms between edge server nodes needs to be considered. This includes the dynamic nature of cluster node selection, inter-node resource complementarity scheduling, and the replacement and fault tolerance of faulty nodes. The aforementioned patents provide a resource allocation method but do not consider the special characteristics of edge server node collaborative clusters and the dynamic elimination strategy for inefficient or faulty nodes within the cluster. This patent, through its proposed task requirement type identification algorithm and resource collaborative allocation strategy, reduces latency while improving resource utilization. Summary of the Invention

[0008] This invention aims to address the problems existing in the prior art and proposes a service-aware resource collaborative allocation method. The technical solution of this invention is as follows:

[0009] A service-aware resource collaborative allocation method includes the following steps:

[0010] S1. Analyze the resource characteristics of edge computing tasks, covering core attributes such as computing requirements, latency requirements, and data volume;

[0011] S2. Perform clustering operations on the task resource characteristics to generate abstract task categories;

[0012] S3. Map task abstraction categories to interpretable business requirement types;

[0013] S4. Based on the type of business needs, candidate collaborative edge node clusters are selected by using the improved adaptive large domain search destruction and repair operators;

[0014] S5. Analyze the benefits and costs of candidate collaborative edge node clusters, and quantify the collaborative benefits of each node combination;

[0015] S6. Based on the results of collaborative benefits, select the edge node combination with the best collaborative benefits as the final collaborative execution cluster.

[0016] Furthermore, in step S1, the resource characteristics of the edge computing task are analyzed, covering core attributes such as computing requirements, latency requirements, and data volume, specifically including:

[0017] S11. Simulate the original description information of the edge computing task, including key parameters such as CPU computing power requirements, memory usage, maximum allowed latency threshold and processing scale;

[0018] S12. Normalize the extracted key parameters to eliminate the influence of different dimensions on subsequent analysis and form a task resource feature vector with a unified dimension.

[0019] Furthermore, in step S2, clustering operations are performed on the task resource features to generate abstract task categories, specifically including:

[0020] S21. A deep embedding clustering algorithm is used to perform unsupervised clustering of task resource features. The deep embedding clustering algorithm is based on an autoencoder model containing an encoder and a decoder. It maps the normalized task resource feature vectors to a low-dimensional latent feature space and performs clustering in the latent feature space to generate several task abstract categories.

[0021] S22. Input the normalized task resource feature vector into the deep model, extract the latent features of the task by training the autoencoder, and perform clustering based on the latent features to obtain several task abstract categories.

[0022] Furthermore, in step S3, mapping the task abstract category to an interpretable business requirement type specifically includes:

[0023] S31. Calculate the relative factor of each task abstract category on the core resource characteristics;

[0024] S32. Based on the relative magnitudes of the factors and combined with knowledge of edge computing business scenarios, abstract categories are mapped to business requirement types.

[0025] Furthermore, in step S4, based on the type of business requirement, and with the help of improved adaptive large-domain search destruction and repair operators, candidate collaborative edge node clusters are selected, specifically including:

[0026] S41. For different business needs, define a disruptive operator for Adaptive Large Neighborhood Search (ALNS): randomly remove a certain proportion of edge nodes from the candidate node pool to simulate scenarios of temporary node failure or insufficient resources.

[0027] S42. Define the repair operator for Adaptive Large Neighborhood Search (ALNS): Combining the expected value of node payoffs in game theory, select the node that maximizes the collaborative potential from the backup node pool and insert it into the damaged node combination; S43. Iteratively execute the damage and repair operations to generate multiple sets of candidate collaborative edge node clusters that meet the constraints of business requirement types.

[0028] Furthermore, in step S5, a game theory analysis is performed on the benefits and costs of the candidate collaborative edge node cluster, and the collaborative benefits of each node combination are quantitatively calculated, specifically including:

[0029] S51. Treat each edge node as a participant in the game, and define the node's "cooperative benefit" as the improvement in task execution efficiency, and the "cooperative cost" as the loss of node resource occupation;

[0030] S52. Construct a game payoff matrix, calculate the payoff function of each node in the candidate node combination, and consider the mutual influence between nodes and resource game.

[0031] S53. Based on the Nash stability theory, solve for the "stable collaborative benefit" of candidate node combinations, which is the overall benefit when each node has no incentive to change its strategy individually.

[0032] Furthermore, in step S6, based on the collaborative benefit results, the optimal combination of edge nodes with the best collaborative benefit is selected, specifically including:

[0033] S61. For each candidate collaborative edge node cluster, extract its stable collaborative revenue value after game theory analysis;

[0034] S62. Compare the stable collaborative benefits of all candidate clusters, and select the node combination with the largest benefit value as the final collaborative execution cluster for the task.

[0035] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements a service-aware resource collaborative allocation method as described in any one of the claims.

[0036] A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a service-aware resource collaborative allocation method as described in any one of the claims.

[0037] The advantages and beneficial effects of this invention are as follows:

[0038] 1. In step S4 of this invention, by leveraging the destruction and repair operators of adaptive large-domain search, combined with the precise matching logic of task business requirement type and edge node resource status, candidate collaborative edge node clusters that meet business requirements can be selected more accurately. Compared with other node selection methods based on random search, the proposed method reduces the number of invalid search iterations and the redundancy of candidate clusters, significantly improving the selection efficiency and accuracy of candidate node clusters.

[0039] 2. In step S5 of this invention, the benefits and costs of candidate collaborative edge node clusters are analyzed by game theory, and the collaborative benefits of each node combination are quantitatively calculated. This solves the problems of uneven resource allocation and insufficient motivation for node collaboration among edge nodes. Furthermore, it can quantify the collaborative value of different node combinations from the perspective of benefits, providing an objective decision-making basis for cluster construction and promoting efficient complementarity and collaborative scheduling of resources among nodes.

[0040] 3. Step S6 of this invention proposes a resource optimization strategy. Based on the results of collaborative benefits, the problem is transformed into an edge node combination selection problem with the goal of maximizing collaborative benefits. This strategy can screen edge node combinations with strong resource complementarity and high collaborative efficiency under the conditions of meeting task computing requirements and latency constraints, significantly improving the overall resource utilization and task processing performance of the edge computing system.

[0041] 4. Compared with existing technologies, the steps proposed in this invention are not simply a combination of conventional algorithms, but rather form a collaborative design at three levels: task requirement identification, collaborative node search, and stability constraints. On the one hand, existing edge computing resource allocation methods typically assume that the task service type is known, or only use a single clustering algorithm to perform coarse-grained task division. There are no solutions that combine deep embedding clustering with entropy weight analysis to automatically mine four types of business requirements—"latency-sensitive, resource-intensive, hybrid, and other"—from a unified task resource feature vector. On the other hand, traditional adaptive large neighborhood search is mostly used for combinatorial optimization problems such as vehicle routing and production scheduling, and rarely applied in conjunction with game theory models to select edge server collaborative clusters. This invention designs destruction and repair operators specifically for business requirement types in ALNS, and introduces stable collaborative benefit calculation based on nash stability, unifying dynamic operator weight updates and the goal of maximizing collaborative benefits under the same framework. Attached Figure Description

[0042] Figure 1 This is a flowchart of the preferred embodiment provided by the present invention;

[0043] Figure 2 A graph for a deep embedding clustering model;

[0044] Figure 3 Here is the algorithm flowchart; Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0046] The technical solution of the present invention to solve the above-mentioned technical problems is:

[0047] 1. Based on the core resource characteristics of edge computing tasks, cluster the task resource characteristics to generate task abstract categories, and map them to interpretable business requirement types.

[0048] 2. For the destruction and repair operators using adaptive large-domain search, candidate collaborative edge node clusters are screened by combining task business requirements with edge node resource status matching logic; then, the benefits and costs of candidate clusters are analyzed based on game theory to quantify the collaborative benefits of each node combination; finally, based on the collaborative benefit results, the edge node combination with the best benefit is selected as the collaborative execution cluster.

[0049] like Figure 1 As shown, the present invention provides a service-aware resource collaborative allocation method, comprising the following steps:

[0050] S1. Analyze the resource characteristics of edge computing tasks, covering core attributes such as computing requirements and latency requirements;

[0051] S2. Perform clustering operations on the task resource characteristics to generate abstract task categories;

[0052] S3. Map task abstraction categories to interpretable business requirement types;

[0053] S4. Based on the type of business needs, candidate collaborative edge node clusters are selected by using the improved adaptive large domain search destruction and repair operators;

[0054] S5. Analyze the benefits and costs of candidate collaborative edge node clusters, and quantify the collaborative benefits of each node combination;

[0055] S6. Based on the results of collaborative benefits, select the edge node combination with the best collaborative benefits as the final collaborative execution cluster.

[0056] In this embodiment, step S1 involves parsing the resource characteristics of the edge computing task, covering core attributes such as computing requirements, latency requirements, and data volume. Specifically, this includes the following steps:

[0057] In this embodiment, in step S1, the edge server constructs a user-content bipartite graph and filters popular content based on the degree of content nodes in the bipartite graph. Specifically, this includes the following steps:

[0058] S11. Simulate the original description information of edge computing tasks, including key parameters such as CPU computing power requirements, memory usage, maximum allowed latency threshold and processing scale of the task, and define the task set and task feature vector;

[0059] Define task set Define the original feature vector of each task t as:

[0060]

[0061] in: CPU computing power requirements (unit: MIPS) Memory usage (unit: MB) This represents the maximum tolerable delay threshold (in seconds). Data processing scale (unit: kB).

[0062] Task type (video, audio, image, text). Network type (5G, 4G).

[0063] S12. Normalize the extracted key parameters to eliminate the influence of different dimensions on subsequent analysis and form a normalized task feature vector. The formula is

[0064]

[0065]

[0066]

[0067] in: For the task The kth original feature, These are the minimum and maximum values ​​of the k-th feature for all tasks, respectively.

[0068] The following example will illustrate the process of classifying task requirements. Assume there are 10 task details as shown below, where... CPU computing power requirements (unit: MIPS) Memory usage (unit: MB) This represents the maximum tolerable delay threshold (in seconds). Data processing scale (unit: kB).

[0069] Task type (video, audio, image, text). The table below shows the network type (5G, 4G):

[0070]

[0071] Normalized task feature vector As shown in the table below:

[0072]

[0073] In this embodiment, step S2, which involves clustering the task resource features and generating abstract task categories, specifically includes the following steps:

[0074] S21. Employ a deep embedding clustering algorithm, which includes a deep model of the encoder and decoder;

[0075] (1) Activation function: ReLU (Modified Linear Unit) is used as the activation function, and the formula is:

[0076]

[0077] (2) Encoder: Normalizes the feature vector Mapping to a low-dimensional latent feature space, the implicit feature associations of the task are extracted using a 3-layer fully connected network:

[0078]

[0079]

[0080]

[0081] in: This is the weight matrix. For bias terms, The activation function is the same as (1).

[0082] (3) Decoder: converts latent features The features are reconstructed as input features, and the effectiveness of feature extraction is verified using a 3-layer fully connected network:

[0083]

[0084]

[0085]

[0086] in: This is the decoder weight matrix. For the bias term, the activation function is the same as (1).

[0087] S22. Input the normalized task resource feature vector into the deep model, extract the latent features of the task by training the autoencoder, and perform clustering based on the latent features to obtain several task abstract categories.

[0088] First, the encoder and decoder are trained with the goal of minimizing the reconstruction loss, which is calculated using the mean squared error (MSE). The data is then input into the clustering layer, the encoder parameters are fixed, and a Gaussian Mixture Model (GMM) is introduced to analyze the latent features. Clustering. GMM assumes that the latent features follow a mixture of K Gaussian distributions, and the probability that the i-th task belongs to the k-th class is:

[0089]

[0090] in: The mixing coefficient of class k The mean Covariance The Gaussian distribution.

[0091] After training, the latent features for each task ,Pick As its abstract category label, we obtain K task abstract categories.

[0092] The following table was obtained:

[0093]

[0094] In this embodiment, step S3, mapping the task abstract category to an interpretable business requirement type, specifically includes the following steps:

[0095] S31. Calculate the entropy weight and comprehensive score of each task abstract category on the core resource features;

[0096] Using "latency attribute" and "resource attribute" as the two main evaluation dimensions, information entropy is used to calculate the "relative factor" of each class in these two dimensions. Classification is achieved through the factor ratio and score threshold. The specific process involves calculating the attribute weight, information entropy, entropy weight, and comprehensive score, as shown in the following formula:

[0097] For the j-th feature in the latency and resource dimensions The attribute weight of the i-th category on this feature , for:

[0098]

[0099] For the j-th feature in the latency and resource dimensions, its information entropy , for:

[0100]

[0101]

[0102] Where: the smaller the entropy value, the higher the distinguishability of the feature between categories; ln(K) is used for normalization, making the entropy value range [0,1].

[0103] For the j-th feature in the latency and resource dimensions, its entropy weight , The larger the weight, the greater its contribution to the score. The formula is:

[0104]

[0105]

[0106] Then the overall score , The formula is:

[0107]

[0108]

[0109] The relative factor is defined as F, and the formula is:

[0110]

[0111] Therefore, the information entropy, entropy weight, and comprehensive score table in the example are as follows:

[0112]

[0113] S32. Based on the relative relationships of the factors and combined with knowledge of edge computing business scenarios, abstract categories are mapped to business requirement types, as shown in the table below:

[0114]

[0115] In this embodiment, step S4 involves selecting candidate collaborative edge node clusters based on the type of business requirement and using adaptive large-domain search's destruction and repair operators. This specifically includes the following steps:

[0116] S41. Define the disruption operator for ALNS: randomly remove a certain proportion of edge nodes from the candidate node pool to simulate scenarios of temporary node failure or resource shortage. Define the objective function for the task type, assuming the current cluster is S and the task type is... Cluster basic benefits For example, as shown in the following formula:

[0117]

[0118] in: This represents the average latency of the cluster; a smaller value indicates better latency performance. This indicates the cluster resource utilization rate; a higher value indicates more efficient resource utilization.

[0119] Indicate the weight of "weaknesses" to ensure that core objectives are prioritized; Weights for mixed types and other types.

[0120] The destruction operator identifies and removes nodes that contribute little to the cluster's task-type revenue by calculating the individual payoff of the objective function for each task type. During the destruction phase, a federation game is employed to remove nodes with low contribution values. The "marginal contribution payment" for cluster S measures the contribution of a node to the revenue of a task type. A smaller value indicates a lower contribution from the node to the revenue of the task type. The formula is:

[0121] The execution process of the destruction operator is as follows: first, calculate the values ​​of all nodes in cluster s. Next, select the first one in ascending order. Nodes The destruction rate is used to obtain the final destroyed cluster. .

[0122] S42. Define the repair operator for ALNS: Combining the estimated value of node payoff in game theory, select the node that maximizes the collaborative potential from the backup node pool and insert it into the broken node combination;

[0123] During the repair process, the repair operator quantifies the expected reward of each task type when a node joins the cluster by analyzing the payoffs of a coalition game, prioritizing the insertion of high-contribution nodes. N represents the complete set of edge nodes, which is first traversed... compute nodes for all subsets T. The marginal benefit of each subset is then weighted by the probability of occurrence of the subset, and the backup nodes are... Adding to the disrupted cluster The expected return is:

[0124]

[0125] The workflow of the repair operator is as follows: first, calculate the backup node pool. All nodes Select the first in descending order 1 node The repair rate is used to obtain the repaired cluster. .

[0126] S43. Iteratively execute destruction and repair operations to generate multiple candidate collaborative edge node clusters that meet the constraints of the business requirement type. Iterate through the destruction and repair operations until the core constraints of the task type or the required number of iterations are met, ultimately generating M candidate clusters that satisfy the constraints. .

[0127] The task classification information is shown in the table below:

[0128]

[0129] against Analysis, assuming the initial candidate node pool contains nodes Damage rate The data for the destruction phase are shown in the table below:

[0130]

[0131] The data from the repair phase is shown in the table below. (Cluster data after repair) :

[0132]

[0133] After the destruction-repair iteration, three candidate clusters were generated, as shown in the table below:

[0134]

[0135] In this embodiment, step S5 involves a game theory analysis of the benefits and costs of the candidate collaborative edge node cluster, quantifying the collaborative benefits of each node combination. This specifically includes the following steps:

[0136] S51. Treat each edge node as a game participant, defining the node's "cooperative benefit" as the improvement in task execution efficiency and the "cooperative cost" as the loss of node resource usage. Calculate the cooperative benefit for each edge node. and collaboration costs The net benefit of a node participating in the cluster is... ,when The nodes have no intention of cooperating.

[0137] Synergistic benefits The definition is as follows:

[0138]

[0139] in: For nodes Latency when performing a task alone; For weights.

[0140] Collaboration costs Defined as:

[0141]

[0142] Where: α and β are weights.

[0143] S52. Calculate the payoff function for each node in the candidate node combination, and define the overall cluster payoff function as follows:

[0144]

[0145] S53. Based on Nash stability theory, solve for the stable cooperative benefit of candidate node combinations, which is the overall benefit when no node has the incentive to change its strategy individually, i.e., the cooperative benefit is greater than or equal to the non-cooperative benefit. For any , must meet The stable collaborative benefit is:

[0146] Synergistic benefits can be derived. Collaboration costs The stable synergistic benefits are shown in the table below:

[0147]

[0148] Design a scoring mechanism for damage / repair operators, dynamically adjust the operator selection probability, and update the weight probability based on the operator score. Define the damage operator score. Repair operator scoring The formula is as follows:

[0149]

[0150]

[0151] in: For indicator functions

[0152] Based on the scores, D represents the set of destruction operators, and R represents the set of repair operators. Weights are updated for all destruction and repair operators; higher scores result in larger weight increments. Initially, initial weights for the destruction and repair operators are set. Define the weight of the destruction operator. Repair operator weights The formula is as follows:

[0153]

[0154]

[0155] in: This is the learning rate.

[0156] Obtain the weights of the destruction operator Repair operator weights Then, the selection probability of the destruction operator is obtained. and the selection probability of the repair operator The formula is as follows:

[0157]

[0158]

[0159] in: Avoid weights of 0

[0160] Then, destroy operator scores, repair operator scores, and destroy operator weights. Repair operator weights The selection probability of the disruptor and the selection probability of the repair operator As shown in the table below:

[0161]

[0162] In this embodiment, step S6, selecting the edge node combination with the optimal collaborative benefit based on the collaborative benefit results, specifically includes the following steps:

[0163] S61. For the M candidate clusters generated in step S4 Calculate the stable collaborative benefit for each candidate collaborative edge node cluster. The formula is as follows:

[0164]

[0165] S62. Compare the stable collaborative benefits of all candidate clusters, and select the node combination with the largest benefit value as the final collaborative execution cluster for the task. The formula is as follows, and the table is as follows:

[0166]

[0167]

[0168] Therefore, regarding the task Select edge nodes Execution yields the greatest benefit.

[0169] This invention addresses the issues of insufficient accuracy and low efficiency in the selection and resource allocation of collaborative edge server clusters in mobile edge networks. It proposes a collaborative edge node cluster determination scheme based on task resource characteristics and game theory. First, the core resource characteristics of edge computing tasks are analyzed, and clustering of these characteristics generates abstract task categories, which are then mapped to interpretable business requirement types. Next, using adaptive large-domain search with destruction and repair operators, and combining the matching logic between task business requirements and edge node resource states, candidate collaborative edge node clusters are selected. Then, game theory analysis is used to analyze the benefits and costs of candidate clusters, quantifying the collaborative benefits of each node combination. Finally, based on the collaborative benefit results, the edge node combination with the optimal benefit is selected as the collaborative execution cluster.

[0170] Compared to other methods, this invention utilizes multi-dimensional clustering based on task resource characteristics to more accurately characterize task heterogeneity and is more suitable for the diverse task types in mobile edge scenarios compared to single clustering algorithms. By employing an improved adaptive large-domain search operator to filter candidate nodes, it reduces invalid searches and candidate cluster redundancy, improving the efficiency and accuracy of node selection. Through game theory quantification of collaborative benefits, it addresses the issues of uneven resource allocation and insufficient collaborative motivation among edge nodes, providing an objective basis for cluster formation. The proposed combined resource optimization strategy can select node combinations with strong resource complementarity while satisfying task constraints, significantly improving the resource utilization and task processing performance of the edge computing system.

[0171] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0172] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0173] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0174] The above embodiments should be understood as illustrative only and not as limiting the scope of protection of the present invention. After reading the description of the present invention, those skilled in the art can make various alterations or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A service-aware resource collaborative allocation method, characterized in that, Includes the following steps: S1. Analyze the resource characteristics of edge computing tasks, covering core attributes such as computing requirements, latency requirements, and data volume; S2. Perform clustering operations on the task resource characteristics to generate abstract task categories; S3. Map task abstraction categories to interpretable business requirement types; S4. Based on the type of business needs, candidate collaborative edge node clusters are selected by using the improved adaptive large domain search destruction and repair operators; S5. Analyze the benefits and costs of candidate collaborative edge node clusters, and quantify the collaborative benefits of each node combination; S6. Based on the results of collaborative benefits, select the edge node combination with the best collaborative benefits as the final collaborative execution cluster.

2. The service-aware resource collaborative allocation method according to claim 1, characterized in that, In step S1, the resource characteristics of the edge computing task are analyzed, covering core attributes such as computing requirements, latency requirements, and data volume, specifically including: S11. Simulate the original description information of the edge computing task, including key parameters such as CPU computing power requirements, memory usage, maximum allowed latency threshold and processing scale; S12. Normalize the extracted key parameters to eliminate the influence of different dimensions on subsequent analysis and form a task resource feature vector with a unified dimension.

3. The service-aware resource collaborative allocation method according to claim 1, characterized in that, In step S2, clustering is performed on the task resource features to generate abstract task categories, specifically including: S21. A deep embedding clustering algorithm is used to perform unsupervised clustering of task resource features. The deep embedding clustering algorithm is based on an autoencoder model containing an encoder and a decoder. It maps the normalized task resource feature vectors to a low-dimensional latent feature space and performs clustering in the latent feature space to generate several task abstract categories. S22. Input the normalized task resource feature vector into the deep model, extract the latent features of the task by training the autoencoder, and perform clustering based on the latent features to obtain several task abstract categories.

4. The service-aware resource collaborative allocation method according to claim 1, characterized in that, In step S3, the task abstraction category is mapped to an interpretable business requirement type, specifically including: S31. Calculate the relative factor of each task abstract category on the core resource characteristics; S32. Based on the relative magnitudes of the factors and combined with knowledge of edge computing business scenarios, abstract categories are mapped to business requirement types.

5. The service-aware resource collaborative allocation method according to claim 1, characterized in that, In step S4, based on the type of business requirement, and with the help of improved adaptive large-domain search destruction and repair operators, candidate collaborative edge node clusters are selected, specifically including: S41. For different business needs, define a disruptive operator for Adaptive Large Neighborhood Search (ALNS): randomly remove a certain proportion of edge nodes from the candidate node pool to simulate scenarios of temporary node failure or insufficient resources. S42. Define the repair operator for the adaptive large neighborhood search ALNS: combining the estimated value of node payoff in game theory, select the node that maximizes the collaborative potential from the backup node pool and insert it into the broken node combination; S43. Iteratively execute damage and repair operations to generate multiple candidate collaborative edge node clusters that meet the constraints of business requirement types.

6. The service-aware resource collaborative allocation method according to claim 1, characterized in that, In step S5, a game theory analysis is performed on the benefits and costs of the candidate collaborative edge node cluster, and the collaborative benefits of each node combination are quantitatively calculated. Specifically, this includes: S51. Treat each edge node as a game participant, define the node's "cooperation benefit" as the improvement in task execution efficiency, and the "cooperation cost" as the loss of node resource usage; S52. Construct a game payoff matrix, calculate the payoff function of each node in the candidate node combination, and consider the mutual influence between nodes and resource game. S53. Based on the Nash stability theory, solve for the "stable collaborative benefit" of candidate node combinations, which is the overall benefit when each node has no incentive to change its strategy individually.

7. The service-aware resource collaborative allocation method according to claim 1, characterized in that, In step S6, based on the collaborative benefit results, the optimal combination of edge nodes with the best collaborative benefit is selected, specifically including: S61. For each candidate collaborative edge node cluster, extract its stable collaborative revenue value after game theory analysis; S62. Compare the stable collaborative benefits of all candidate clusters, and select the node combination with the largest benefit value as the final collaborative execution cluster for the task.

8. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the service-aware resource collaborative allocation method as described in any one of claims 1 to 7.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the service-aware resource collaborative allocation method as described in any one of claims 1 to 7.

Images