A method for optimizing task precoding and resource allocation in wireless edge computing networks

Abstract

This invention provides a method for optimizing task precoding and resource allocation in wireless edge computing networks. The method includes: constructing a wireless edge computing network architecture model, establishing task execution latency and energy consumption models, and establishing a joint optimization problem for computing power scheduling and resource allocation in the computing network with the goal of minimizing task execution latency. Further, the joint optimization problem is decomposed into a task precoding sub-problem and joint task partitioning, computing power association, and computing power allocation sub-problems. Through a joint design based on convex optimization and a multi-agent reinforcement learning algorithm, the method effectively addresses challenges such as communication overhead in task transmission, underutilization of computing power, and task privacy and security in dynamic and complex environments. This invention's method can effectively guide agents to obtain optimal strategies for task precoding, computing power association, and resource optimization at a faster convergence speed while ensuring efficiency and stable performance.

Description

Technical Field

[0001] This invention relates to the field of edge computing network technology, and in particular to a method for task precoding and resource allocation optimization in wireless edge computing networks. Background Technology

[0002] The advent of the digital age has driven the rapid development of compute-intensive wireless services, leading to a continuous increase in demand for computing resources and posing significant challenges to traditional computing architectures. Computing Power Network (CPN) has been proposed as a promising new architecture that can effectively address the explosive growth in computing power demand. Through ubiquitous network connectivity, CPN integrates multi-level computing power and storage resources, connecting geographically dispersed computing resources, coordinating the allocation and scheduling of computing tasks, and providing users with on-demand, comprehensive computing services based on the deep integration of communication, computing, and network resources.

[0003] The construction of computing networks relies on edge intelligence as a key supporting technology. Edge intelligence introduces computing and caching resources to the edge of mobile networks, deploys artificial intelligence services on the edge network, and establishes dynamic and adaptive edge network management and maintenance to significantly reduce communication overhead and latency, providing diverse, distributed, low-latency, and reliable intelligent services. The concept of edge intelligence includes the following two aspects: First, deploying artificial intelligence services through edge computing to achieve the combination of edge and intelligence; second, integrating artificial intelligence technology into the edge computing network to establish dynamic and adaptive edge network management and maintenance, forming an integration of intelligence and edge technologies.

[0004] By integrating edge intelligence with computing networks, computing networks can provide flexible scheduling and optimization of computing resources for edge intelligence. Simultaneously, edge intelligence expands computing services and enables intelligent deployment of computing networks, achieving unparalleled advancements in processing power and real-time decision-making. The enhanced overall scheduling capabilities of computing networks and the enhanced computing performance of MEC (Mobile Edge Computing) have also spurred a series of computationally intensive innovative applications such as autonomous driving, VR (Virtual Reality) / AR (Augmented Reality), and smart cities. Beyond these application characteristics, CPN still holds significant advantages from an implementation and deployment perspective. Any terminal device with computing power in future mobile communication systems can be integrated into the edge network as a computing node, providing immense flexibility and compatibility through the optimized and efficient utilization of edge networks and computing resources.

[0005] With the development of computing networks and edge networks, the joint optimization of computing power scheduling and resource allocation has become a hot research topic, presenting considerable complexity and challenges. However, considering the large number of heterogeneous and ubiquitous computing nodes in computing power networks, how to efficiently utilize, match, and deploy computing resources in edge networks is a pressing issue. Advances in existing computing paradigms such as edge computing have provided diverse solutions, but they have failed to address current pain points, including high communication overhead in computing power scheduling, underutilization of computing resources, privacy and security of task data, and the low-latency requirements of intelligent tasks. Computing power is task-oriented; for specific computing tasks, the scheduling and networking strategies need to be adjusted according to the requirements and complexity of the task. In complex environments such as the high mobility, limited resources, and time-varying nature of mobile networks, the heterogeneous resources of different devices exacerbate resource shortages. In real-world scenarios (such as vehicular networks), the probability of dynamic time-varying channel states and unstable network topologies is high. Implementing computing power scheduling inevitably leads to high communication overhead, potentially facing the dilemma of excessive upload traffic and unacceptable upload times. The limited computing resources of computing nodes may be insufficient to meet the diverse needs of the network. More seriously, the coordination of distributed computing resources during the computing power scheduling process greatly increases the likelihood of leaking task data, leading to privacy threats.

[0006] Furthermore, the environment in computing networks is complex and variable, and existing resource optimization schemes are not suitable for computing network architectures. The interaction between distributed computing nodes increases the randomness and complexity of algorithms. The enhanced requirements of 6G applications make existing computation offloading methods unsuitable for latency performance. The limited resources of heterogeneous devices and time-varying channel environments require flexible scheduling and elastic orchestration of ubiquitous computing resources, making joint optimization even more complex. Summary of the Invention

[0007] Embodiments of the present invention provide a method for optimizing task precoding and resource allocation in a wireless edge computing network, so as to achieve joint optimization of computing power scheduling and resource allocation, minimize task execution latency, and improve the convergence speed and efficiency of learning.

[0008] To achieve the above objectives, the present invention adopts the following technical solution.

[0009] A method for optimizing task precoding and resource allocation in a wireless edge computing network includes:

[0010] Construct a wireless edge computing network architecture model;

[0011] Based on the aforementioned wireless edge computing network architecture model, a joint optimization problem of computing power scheduling and resource allocation in the wireless edge computing network with the goal of minimizing task execution latency is established.

[0012] A two-layer optimization algorithm is designed to decompose the joint optimization problem into a task precoding subproblem and a joint task partitioning, computing power association, and computing power allocation subproblem.

[0013] After equivalence analysis of the task precoding subproblem, a precoding strategy is obtained by designing a convex optimization-based algorithm. A multi-agent deep reinforcement learning algorithm is designed to solve the joint task partitioning, computing power association, and computing power allocation subproblems, thereby obtaining optimized strategies for task partitioning, computing power association, and resource allocation.

[0014] Preferably, the construction of the wireless edge computing network system architecture model, initialization of the task nodes, computing nodes, and computing power scheduling system environment in the network, and establishment of task execution latency and energy consumption models include:

[0015] A Wireless Edge Computing Network (WECPN) architecture is constructed, comprising a wireless edge access layer, a computing power adaptation layer, and an application layer. In the wireless edge access layer, nodes that generate computing tasks are designated as task nodes. In the computing power adaptation layer, computing power nodes that utilize computing resources for computation and communication are configured. In the WECPN architecture, it is assumed that there are N heterogeneous, ubiquitous computing power nodes, represented by a set of computing power nodes. This represents the set of computing nodes, where N = K + M. Divided into terminal service device computing node sets and server computing node set

[0016] Assume the indices of the computing power node and the task node are represented by i and n respectively, and the state of the computing power node is represented as... in χ represents the current computing power and CPU frequency of computing node i, where f and χ represent the current computing power and CPU frequency of computing node i. i Indicates the position of computing node i; sets the triple Γ n = <D n , ψ n , τ n > represents the generated computational task, where D n It calculates the task size, ψ n It is computational density, that is, the number of CPU cycles required to complete a 1-bit computation task, τ. n This represents the maximum tolerable delay for completing the computation task.

[0017] Preferably, the joint optimization problem of computing power scheduling and resource allocation in the wireless edge computing power network with the goal of minimizing task execution latency, based on the wireless edge computing power network system architecture model, includes:

[0018] The task execution latency in a wireless edge computing network includes task precoding overhead, task computation, and communication latency overhead. The average task representation rate of task n is expressed as ε. n , ε n ∈(0,1], the size of the task processed by the user in the t-th time slot during the precoding stage is expressed as: For each user u n The computation time for extracting the task representation from the original task n is:

[0019]

[0020] in This is the computational cost required to extract the task representation for user u. n The execution performance of the task and the extraction rate of the task expression are modeled as follows:

[0021]

[0022] When local computing mode is selected, the computing subtasks will be performed on the terminal device u. n Execution occurs locally, and the total local execution time is the computation time. Task u is executed within time slot t. n The estimated time required is expressed as follows:

[0023]

[0024] When the service device is selected for computation, task u will be executed. n The subtask is offloaded to the service device u in time slot t via wireless communication. d , And execute remotely;

[0025] u n -u d Uplink transmission: in

[0026] u d calculate:

[0027] u d -u n Downlink transmission:

[0028] The total task execution time of the service device computing power node is expressed as:

[0029]

[0030] Assume that there is no overlap between any subtasks that satisfy the condition, i.e., satisfy the condition.

[0031] u n -u s Uplink transmission: in

[0032] u s calculate:

[0033] The collaborative computation time of server nodes is expressed as:

[0034]

[0035] T n The task execution latency corresponding to the computation task is expressed as:

[0036]

[0037] Computational energy consumption for performing the task: The energy consumption of all computing nodes executing the entire computation task is then...

[0038] Communication power consumption: The energy consumption of non-link task transmission across all computing nodes is expressed as:

[0039] Total energy consumption:

[0040] The joint optimization problem P1 of computing power scheduling and resource allocation in a wireless edge computing network, with the goal of minimizing task execution latency, is as follows:

[0041]

[0042] Preferably, the method of decomposing the joint optimization problem into a task precoding subproblem and a joint task partitioning, computational power association, and computational power allocation subproblem using a two-layer optimization algorithm includes:

[0043] A two-layer optimization algorithm is designed to decompose the joint optimization problem into a task precoding subproblem and a joint task partitioning, computing power association, and computing power allocation subproblem.

[0044] Given the task partitioning, computing power association, and computing power allocation, the task precoding subproblem P2 can be expressed as:

[0045]

[0046] The sub-problem P3 concerning joint task partitioning, computing power association, and computing power allocation is expressed as follows:

[0047]

[0048] Preferably, the precoding strategy is obtained by designing a convex optimization-based algorithm after equivalence analysis of the task precoding subproblem, including:

[0049] The changing trends of task performance and computational complexity functions are analyzed, and expressions relating computational complexity and computational performance to task expression rate are given. For additional computational complexity and task performance functions, a fitting algorithm is used to solve them. An equivalence analysis is performed on the task precoding subproblem, transforming it into a convex optimization problem. The Karl von Sort-Kuhn-Tucker KKT conditions are used to solve the convex optimization problem, yielding the optimal solution ε of the precoding strategy, i.e., the task expression rate.

[0050] Preferably, the step of solving the joint task partitioning, computing power association, and computing power allocation subproblems by designing a multi-agent deep reinforcement learning algorithm to obtain optimization strategies for task partitioning, computing power association, and resource allocation includes:

[0051] The joint task partitioning, computing power association, and computing power allocation sub-problems P3 are transformed into Markov decision processes, and the state space, action space, and reward function of the multi-agent reinforcement learning algorithm are established.

[0052] (1) Determine the state space;

[0053] In each time slot, the agent allocates tasks and computing resources. This represents the joint observation space of all agents, where each agent obtains a portion of the observations from the environment. n (t) is represented as χ n (t), f i (t), d n (t), h n (t)}, The global state of the environment is represented in each time slot as:

[0054] This represents the joint action space of all agents. Each agent needs to make decisions regarding task partitioning, computational power association, and computational power allocation. The actions of an agent in a time slot are represented as follows: All agent actions are represented as

[0055] The reward function is designed as a hybrid reward function consistent with the constraints and optimization objectives;

[0056] The joint task partitioning, computing power association, and computing power allocation subproblems are solved by a multi-agent deep reinforcement learning algorithm to obtain the task partitioning, computing power association, and resource allocation strategies, namely, which computing power nodes a task is associated with, the proportion of computing power nodes that execute tasks, and the computing resources allocated when computing power resources are used to execute tasks.

[0057] Preferably, the reward function includes: immediate reward and long-term reward. The immediate reward is the reward for evaluating the result after each task is completed, involving a hybrid reward that is consistent with the constraints and optimization objectives. The long-term reward uses a discount factor to balance between the immediate reward and the long-term reward. The agent continuously interacts with the environment to maximize the long-term discounted reward.

[0058] State normalization: State normalization is used to scale the states, placing them within the range of [0, 1].

[0059] Action mask: Constrain all continuous variables such as task partitioning, computing power association, and computing power allocation to [0, 1], and take the values ​​of task partitioning variables from the set (0, 0.01, 0.02, ..., 1);

[0060] The total reward for all computing task nodes is:

[0061]

[0062] The long-term cumulative reward R is calculated as follows:

[0063]

[0064] As can be seen from the technical solutions provided by the embodiments of the present invention above, the method of the present invention integrates computing power networks with edge intelligence. By transforming computing power scheduling into task migration, it proposes an edge intelligence-driven, task-aware wireless edge computing power network model to support users with limited available resources to collaboratively execute various intelligent tasks through ubiquitous computing power nodes. The two-layer optimized task precoding and resource allocation algorithm proposed in this invention can effectively guide the agent to approximate the optimal strategy for computing power scheduling and resource optimization while ensuring efficiency and stable performance, thereby obtaining the optimized solution at a faster convergence speed.

[0065] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and will become apparent from the description or may be learned by practice of the invention. Attached Figure Description

[0066] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 A flowchart illustrating a task precoding and resource allocation optimization design method in a wireless edge computing network, provided as an embodiment of the present invention;

[0068] Figure 2 This is a schematic diagram of the architecture of a wireless edge computing network provided in an embodiment of the present invention. Detailed Implementation

[0069] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0070] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or couplings. The term “and / or” as used herein includes any and all combinations of one or more of the associated listed items.

[0071] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0072] To facilitate understanding of the embodiments of the present invention, the following will provide further explanation and description with reference to the accompanying drawings and several specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0073] This invention considers edge-intelligent driven wireless computing networks and proposes a two-layer optimized task precoding and resource optimization algorithm to support users with limited available resources in collaboratively executing various intelligent tasks through ubiquitous computing nodes. Addressing the challenges posed by channel uncertainty and resource finiteness, a task precoding scheme and task partitioning, computing power association, and computing power resource allocation scheme are jointly designed. At the computing power task nodes, a semantically aware task precoding method is proposed to obtain the semantic representation of the task. Based on this, a multi-agent reinforcement learning algorithm is further proposed to dynamically and jointly design and arrange task partitioning, computing power association, and resource allocation, enhancing system efficiency and robustness while minimizing system latency overhead.

[0074] This invention proposes a two-layer optimization algorithm to jointly optimize task precoding, computing power task partitioning, computing power association, and resource allocation in wireless edge computing networks, aiming to minimize the task execution latency of all computing power nodes under limited resource and task performance constraints. In the inner layer, a convex optimization-based method is used to derive the optimal task representation rate using a closed-form expression. In the outer layer, based on the obtained optimal task semantic representation, a multi-agent deep reinforcement learning algorithm is proposed to efficiently solve the joint optimization subproblem. The proposed algorithm aims to achieve fast convergence while effectively shortening task execution time, enhancing system efficiency and robustness.

[0075] The flowchart of a task precoding and resource allocation optimization method in a wireless edge computing network provided by this invention is as follows: Figure 1 As shown, the processing steps include the following;

[0076] Step S10: Construct a wireless edge computing network architecture model.

[0077] Step S20: Based on the above wireless edge computing network system architecture model, establish a joint optimization problem of computing power scheduling and resource allocation in the wireless edge computing network with the goal of minimizing task execution latency.

[0078] Step S30: Use a two-layer optimization algorithm to decompose the above joint optimization problem into a task precoding subproblem, a joint task partitioning subproblem, a computing power association subproblem, and a computing power allocation optimization subproblem.

[0079] Step S40: After performing equivalence analysis on the task precoding subproblem, a precoding strategy is obtained by designing an algorithm based on convex optimization; the above joint task partitioning, computing power association, and computing power allocation optimization subproblems are solved by an algorithm based on multi-agent deep reinforcement learning to obtain the task partitioning, computing power association, and resource allocation strategies.

[0080] The present invention proposes a wireless edge computing network model with the following structure: Figure 2As shown, a two-layer optimization algorithm is proposed for task precoding and resource allocation optimization in wireless edge computing networks. This algorithm achieves joint optimization of task precoding, task partitioning, computing power association, and resource allocation to minimize task execution latency. Through a joint design based on convex optimization and a multi-agent reinforcement learning algorithm, this algorithm effectively addresses challenges such as high communication overhead for task transmission, underutilization of computing power, and privacy and security of computing power task data in dynamic and complex environments. Furthermore, it effectively guides agents to obtain optimal strategies for task precoding, computing power association, and resource optimization, achieving faster convergence and obtaining the optimal solution.

[0081] The above step S10 specifically includes:

[0082] Step 1: Construct a wireless edge computing network system architecture model.

[0083] We construct a WECPN (Wireless Edge Computing Power Networks) network architecture consisting of a wireless edge access layer, a computing power adaptation layer, and an application layer. Figure 2 This is a schematic diagram of a WECPN architecture provided in an embodiment of the present invention. At the wireless edge access layer, some users need to perform computing tasks, i.e., computing task nodes. Therefore, users connect to nearby edge servers via wireless communication. Computing resources are allocated according to demand using the proposed learning algorithm. The computing power adaptation layer is deployed at the network edge, including interconnected heterogeneous and ubiquitous computing power nodes, such as service devices, edge servers, and cloud servers. The computing power adaptation layer utilizes computing resources to promote the integration of the entire network, enhance the computing capabilities of nodes, and promote proactive, intelligent integration and collaboration between nodes. The application layer provides support for applications such as intelligent transportation, VR, remote interactive services, and big data AI services. Unlike traditional vertically scheduled MEC systems, WECPN is based on ubiquitous algorithm resources, forming a computing resource pool for heterogeneous and geographically distributed computing resources such as terminals, edge computing servers, and cloud servers, thereby coordinating and scheduling service requests and comprehensively determining the computing and communication between ubiquitous computing nodes.

[0084] In WECPN, assuming there are N heterogeneous, ubiquitous nodes, a set is used. This is represented as N = K + M. Specifically, the computing power node set can be divided into the computing power of terminal service devices. and server computing power All have the computing capabilities to execute computational tasks. Through the collaborative adaptation of computing resources, WECPN provides on-demand and efficient computing services for various applications. Note that the terminal device can be both a task initiator and a task executor. Assume the indices of the computing power node and the task node are represented as i and n, respectively. Furthermore, the state of the computing power node is represented as... in and f i This represents the current computing power and CPU frequency of computing node i. χ i This indicates the location of the power node i. Furthermore, let the triple Γ be... n = <D n , ψ n , τ n > represents the user's computational task, where D n It calculates the task size, ψ n It is computational density, that is, the number of CPU cycles required to complete a 1-bit computation task, τ. n This represents the maximum tolerable latency for completing the computation task. In real-world scenarios, computing power is difficult to schedule directly. This invention considers transforming computing power scheduling into task migration, while the computation task can be decomposed into multiple sub-tasks for execution to make fuller use of heterogeneous computing resources.

[0085] Step 2: Establish the latency model and energy consumption model for task execution.

[0086] Step 2.1 Establishing the Delay Model

[0087] This section will analyze system latency in detail, including task precoding overhead, task computation, and communication latency overhead.

[0088] Task-aware precoding

[0089] Let ε represent the average task expression rate of task n. n , ε n ∈(0,1). The size of the task processed by the user in the t-th time slot during the task-aware precoding phase is represented as: As mentioned above, to reduce transmission overhead, each device equipped with a task precoding processing unit extracts task representation information instead of the raw data. Note that each user performing the extraction incurs an additional workload for the extraction process, for each user u n The computation time for extracting the task representation from the original task n is:

[0090]

[0091] in It is the computational cost required to extract the task representation, a function representing the task representation rate. For user u n The execution performance of the task is expressed as:

[0092]

[0093] Local computing time

[0094] When local computing mode is selected, the computing subtasks will be performed on the terminal device u. n Execute locally. The total local execution time is the computation time. Execute task u in time slot t. n The estimated time required is expressed as follows:

[0095]

[0096] in This indicates the ratio of subtasks assigned to itself.

[0097] Service equipment calculation

[0098] When selecting a service device for calculation, u n The subtask is offloaded to the service device u in time slot t via wireless communication. d , And execute remotely.

[0099] u n -u d Uplink transmission: in B nd (t) and P nd (t) represents the bandwidth and transmission power allocated to time slot t, h nd For user u n With service equipment u d The current channel coefficients between. This indicates the proportion of subtasks allocated to the service device.

[0100] u d calculate:

[0101] u d -u n Downlink transmission:

[0102] In summary, the total execution time of the service device's computing power nodes can be expressed as:

[0103]

[0104] Server node computation

[0105] Due to resource and latency limitations, users may further migrate computational tasks to servers with more powerful computing capabilities. For ease of discussion, we assume that there is no overlap between any subtasks that satisfy the condition, i.e., satisfying the condition.

[0106] u n -us Uplink transmission: in h ns (t) represents user u n With server u s The current channel coefficients between.

[0107] u s calculate:

[0108] In this case, the collaborative computation time can be expressed as:

[0109]

[0110] T ni The overall execution time for the computation task is expressed as:

[0111]

[0112] Step 2.2: Energy Consumption Model Establishment

[0113] This section mainly covers the components of computing and communication power consumption, as detailed below:

[0114] Calculate energy consumption: Then, the energy consumption of all computing nodes executing the entire computation task is...

[0115] Communication power consumption: The energy consumption of off-link task transmission across all computing nodes can be expressed as:

[0116] Total energy consumption:

[0117] The above step S20 specifically includes:

[0118] The problem of minimizing system latency was established.

[0119] This invention considers resource constraints, energy consumption constraints, and task performance constraints, and minimizes task completion time by jointly optimizing task precoding, task partitioning, computing power partitioning, and computing power resource allocation, which can be expressed as:

[0120]

[0121] The above step S30 specifically includes:

[0122] Due to the presence of binary variables and the coupling of multidimensional optimization variables, this optimization problem can be identified as a mixed-integer programming problem. The heterogeneity of computing power nodes and the time-varying nature of the environment result in an extremely large computational burden for the optimization model, making task transfer and computation a challenge. A direct solution is often difficult to devise. To improve processing efficiency, a two-layer optimization algorithm framework is proposed. In the inner algorithm, given the remaining variables, a convex optimization method is used to optimize the task representation rate ε. n Then, in the outer layer algorithm, a multi-agent reinforcement learning algorithm is designed using the obtained optimal task representation rate to optimize the remaining variables A, Y, and F.

[0123] 1) Inner layer: Optimize computing power task precoding:

[0124] Given the task partitioning, computing power association, and computing power allocation, the corresponding computing power task-aware precoding problem can be formulated as follows:

[0125]

[0126] 2) Outer layer: Sub-problems of joint task partitioning, computational power correlation, and computational power allocation:

[0127] For a given task-aware representation extraction, problem P1 can be simplified to:

[0128]

[0129] The above step S40 specifically includes:

[0130] Step 1: Solving the task precoding subproblem

[0131] The key challenge in solving P2 is the computational complexity of task representation extraction. and task execution performance The implicit expression for the additional computation introduced by the task expression rate is discussed. To address this issue, the changing trends of the task performance and computational cost functions are first analyzed, and an expression for the computational cost is given. For task performance, a corresponding expression is provided. For the additional computational cost and task performance function, a fitting algorithm can be used to solve the problem. Furthermore, an equivalence analysis is performed on the problem, transforming it into a convex optimization problem. The transformed convex problem is solved using KKT (Karush-Kuhn-Tucker) conditions to obtain the optimal solution ε for the precoding strategy, i.e., the task expression rate. The specific solution steps are as follows:

[0132] 1. Define the Lagrange multipliers to obtain the corresponding Lagrange functions;

[0133] 2. Based on the Lagrangian function, the first-order optimality condition, the original feasibility condition, the dual feasibility condition, and the steady-state condition are obtained;

[0134] 3. Combining the above original feasible conditions, steady-state conditions, and first-order optimality conditions, the optimal precoding strategy, i.e., the ratio of the amount of task extracted before the task is transmitted to the computing power node to the original task size, can yield the optimal solution.

[0135] Step 2: Solving the sub-problems of joint task partitioning, computing power association, and computing power allocation optimization.

[0136] Considering the coupling relationships between the subproblems defined in P3 and the lack of a central controller, a multi-agent reinforcement learning algorithm is employed to solve the problem efficiently and accurately. Therefore, problem P3 is formulated as an MDP that maximizes the cumulative discount reward.

[0137] State space: Represents the global state of the environment, expressed in each time slot as:

[0138] Observation space: Represents the joint observation space of all agents, where each agent obtains a portion of the observations o from the environment. n (t) can be represented as Where Γ n (t) and χ n (t) represents the intrinsic parameters associated with each agent, while the rest are perceived parameters, representing the local parameters that each agent collects from the environment. Based on this, the following conditions are met.

[0139] Action space: This represents the joint action space of all agents. Based on local observations, each agent needs to make decisions regarding task partitioning, computational power association, and computational power allocation. An agent's actions within a time slot can be represented as... All agent actions are represented as

[0140] Rewards: The goal of each agent is to learn an optimal policy, effectively utilize computing resources, reduce task execution latency, and ensure successful task execution. This invention designs the reward as a hybrid reward function consistent with constraints and optimization objectives. It consists of immediate and long-term rewards. The immediate reward is the reward for evaluating the result after each task is completed, involving a hybrid reward consistent with constraints and optimization objectives. The long-term reward uses a discount factor to balance between immediate and long-term rewards. The agent continuously interacts with the environment to maximize the long-term discounted reward, effectively addressing the sparsity challenge observed in single-layer reward functions based solely on the final objective, ensuring fast and stable convergence during training.

[0141] In addition, state normalization and action masking were used to improve the efficiency of the algorithm and adapt the output to our formulaic problem.

[0142] ①State normalization: In order to speed up convergence, enhance training stability, and make the policy have greater generalization ability,

[0143] State normalization is used to scale the values ​​in the state space, placing them within the range of [0, 1].

[0144] ② Action Mask: To reduce the dimensionality of the action space, all continuous variables, including task partitioning, computational power association, and computational power allocation, are constrained to [0, 1]. Furthermore, considering that tasks are not arbitrarily separable in real-world scenarios, It can only take a limited number of values. At the same time, to reduce the dimensionality of the action space, for the task partitioning variable, we consider it to originate from the set (0, 0.01, 0.02, ...). … , 1) takes the value.

[0145] Finally, the total reward for all computation task nodes is:

[0146]

[0147] The long-term cumulative reward R is calculated as follows:

[0148]

[0149] The designed multi-agent deep reinforcement learning algorithm is used to solve the joint task partitioning, computing power association, and computing power allocation subproblems, thereby obtaining the task partitioning, computing power association, and resource allocation strategies, namely, which computing power nodes a task is associated with, the proportion of computing power nodes that execute tasks, and the computing resources allocated when computing power resources are used to execute tasks.

[0150] In summary, this invention addresses pain points such as high communication overhead in task transmission, underutilization of computing power, privacy and security of computing power task data, and low latency requirements for intelligent tasks. It proposes a computing power framework based on edge intelligence—a wireless edge computing network model. Under resource constraints and task performance limitations, it minimizes task execution latency by jointly optimizing task precoding, computing power association, task partitioning, and resource allocation. Considering the complexity of the problem and the dynamic nature of the system, the problem is decomposed into a task precoding subproblem and a joint task partitioning, computing power association, and resource allocation subproblem. Then, a two-layer optimization algorithm is proposed to formalize the problem. Specifically, in the inner layer, a semantically aware task precoding method is proposed, using a convex optimization-based approach to derive the optimal task representation rate using a closed-form expression. In the outer layer, based on the obtained optimal task representation rate, a multi-agent deep reinforcement learning algorithm is proposed to efficiently solve the joint optimization subproblem.

[0151] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0152] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

[0153] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for apparatus or system embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The apparatus and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0154] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for optimizing task precoding and resource allocation in a wireless edge computing network, characterized in that, The application relates to a wireless edge computing power network system architecture model, a joint optimization problem of computing power scheduling and resource allocation in a wireless edge computing power network with a task execution time delay minimization target based on the wireless edge computing power network system architecture model, a double-layer optimization algorithm used for decomposing the joint optimization problem into a task pre-coding sub-problem and a joint task division, computing power association and computing power allocation sub-problem, an equivalent analysis of the task pre-coding sub-problem, a pre-coding strategy obtained by designing a convex optimization-based algorithm, a multi-agent deep reinforcement learning algorithm used for solving the joint task division, computing power association and computing power allocation sub-problem, and an optimization strategy of task division, computing power association and resource allocation. The application relates to a wireless edge computing power network system architecture model, a joint optimization problem of computing power scheduling and resource allocation in a wireless edge computing power network with a task execution time delay minimization target based on the wireless edge computing power network system architecture model, a double-layer optimization algorithm used for decomposing the joint optimization problem into a task pre-coding sub-problem and a joint task division, computing power association and computing power allocation sub-problem, an equivalent analysis of the task pre-coding sub-problem, a pre-coding strategy obtained by designing a convex optimization-based algorithm, a multi-agent deep reinforcement learning algorithm used for solving the joint task division, computing power association and computing power allocation sub-problem, and an optimization strategy of task division, computing power association and resource allocation. The application relates to a wireless edge computing power network system architecture model, a joint optimization problem of computing power scheduling and resource allocation in a wireless edge computing power network with a task execution time delay minimization target based on the wireless edge computing power network system architecture model, a double-layer optimization algorithm used for decomposing the joint optimization problem into a task pre-coding sub-problem and a joint task division, computing power association and computing power allocation sub-problem, an equivalent analysis of the task pre-coding sub-problem, a pre-coding strategy obtained by designing a convex optimization-based algorithm, a multi-agent deep reinforcement learning algorithm used for solving the joint task division, computing power association and computing power allocation sub-problem, and an optimization strategy of task division, computing power association and resource allocation. The application relates to a wireless edge computing power network system architecture model, a joint optimization problem of computing power scheduling and resource allocation in a wireless edge computing power network with a task execution time delay minimization target based on the wireless edge computing power network system architecture model, a double-layer optimization algorithm used for decomposing the joint optimization problem into a task pre-coding sub-problem and a joint task division, computing power association and computing power allocation sub-problem, an equivalent analysis of the task pre-coding sub-problem, a pre-coding strategy obtained by designing a convex optimization-based algorithm, a multi-agent deep reinforcement learning algorithm used for solving the joint task division, computing power association and computing power allocation sub-problem, and an optimization strategy of task division, computing power association and resource allocation. The application relates to a wireless edge computing power network system architecture model, a joint optimization problem of computing power scheduling and resource allocation in a wireless edge computing power network with a task execution time delay minimization target based on the wireless edge computing power network system architecture model, a double-layer optimization algorithm used for decomposing the joint optimization problem into a task pre-coding sub-problem and a joint task division, computing power association and computing power allocation sub-problem, an equivalent analysis of the task pre-coding sub-problem, a pre-coding strategy obtained by designing a convex optimization-based algorithm, a multi-agent deep reinforcement learning algorithm used for solving the joint task division, computing power association and computing power allocation sub-problem, and an optimization strategy of task division, computing power association and resource allocation. The application relates to a wireless edge computing power network system architecture model, a joint optimization problem of computing power scheduling and resource allocation in a wireless edge computing power network with a task execution time delay minimization target based on the wireless edge computing power network system architecture model, a double-layer optimization algorithm used for decomposing the joint optimization problem into a task pre-coding sub-problem and a joint task division, computing power association and computing power allocation sub-problem, an equivalent analysis of the task pre-coding sub-problem, a pre-coding strategy obtained by designing a convex optimization-based algorithm, a multi-agent deep reinforcement learning algorithm used for solving the joint task division, computing power association and computing power allocation sub-problem, and an optimization strategy of task division, computing power association and resource allocation. The application relates to a wireless edge computing power network system architecture model, a joint optimization problem of computing power scheduling and resource allocation in a wireless edge computing power network with a task execution time delay minimization target based on the wireless edge computing power network system architecture model, a double-layer optimization algorithm used for decomposing the joint optimization problem into a task pre-coding sub-problem and a joint task division, computing power association and computing power allocation sub-problem, an equivalent analysis of the task pre-coding sub-problem, a pre-coding strategy obtained by designing a convex optimization-based algorithm, a multi-agent deep reinforcement learning algorithm used for solving the joint task division, computing power association and computing power allocation sub-problem, and an optimization strategy of task division, computing power association and resource allocation. The application relates to a wireless edge computing power network system architecture model, a joint optimization problem of computing power scheduling and resource allocation in a wireless edge computing power network with a task execution time delay minimization target based on the wireless edge computing power network system architecture model, a double-layer optimization algorithm used for decomposing the joint optimization problem into a task pre-coding sub-problem and a joint task division, computing power association and computing power allocation sub-problem, an equivalent analysis of the task pre-coding sub-problem, a pre-coding strategy obtained by designing a convex optimization-based algorithm, a multi-agent deep reinforcement learning algorithm used for solving the joint task division, computing power association and computing power allocation sub-problem, and an optimization strategy of task division, computing power association and resource allocation. ​ ​ The objective functions of the above formula (8) and formula (9) represent the minimization of the execution time T of all tasks n , ε, A, Y, F are optimization variables, respectively represented as task expression rate, task association, task partitioning and computing power allocation variables; wherein is a binary variable, indicates that the task node is associated with the computing power node, otherwise The constraint condition ε n ∈(0, 1] represents the average task expression rate range of the task node, indicates the execution performance of the task needs to be greater than the threshold value indicates that the total energy consumption overhead of the computing power node cannot exceed its maximum energy budget indicates that there is no overlap between all subtasks of the task node, represents the proportion of tasks executed locally by the task node, is the proportion of tasks allocated by the task node to all service devices and, is the proportion of tasks allocated by the task node u n to the service device u d , is the proportion of tasks allocated by the task node to all server computing nodes and, is the proportion of tasks allocated by the task node u n to the server computing node u s , represents the range limit of task partitioning; indicates that only when the task is associated with the computing power node can the task be allocated; represents the computing capacity range of the node; ​ ​ ​ In each time slot, the agents partition the task and computing resources, denotes the joint observation space of all agents, each agent gets a partial observation value o from the environment n (t) denotes the denotes the global state of the environment, is the set of distances between task nodes and computing nodes, is the set of channel gains between task nodes and computing nodes, denoted as The joint action space of all agents, each agent needs to make decisions on task partition, computing power association and computing power allocation, the action of agent at time slot is denoted as The action of all agents is denoted as ​ ​ 2. The method of claim 1, wherein, ​ A wireless edge computing power network (WECPN) architecture including a wireless edge access layer, a computing power adaptation layer, and an application layer is constructed. A node generating a computing task is set as a task node in the wireless edge access layer. A computing power node using computing power resources for computing and communication is set in the computing power adaptation layer. In the WECPN architecture, it is assumed that there are N heterogeneous and ubiquitous computing power nodes, and a computing power node set is represented as follows: wherein N=K+M, the computing power node set is divided into a terminal service device computing node set and a server computing node set Assuming the index of the computing node and the task node are represented as i and n respectively, the state of the computing node is represented as where ζ(f i ) and f i represent the current computing power and CPU computing frequency of the computing node i, χ i represents the location of the computing node i; setting a triple Γ n = <D n , ψ n , τ n > represents the generated computing task, where D n is the computing task size, ψ n is the computing density, i.e. the number of CPU cycles required to complete a 1-bit computing task, and τ n represents the maximum tolerable delay for completing the computing task.

3. The method of claim 2, wherein, ​ The task execution delay in the wireless edge computing network includes task pre-encoding overhead, task computing and communication delay overhead, and the average task representation rate of task n is represented as ε n , ε n ∈(0, 1], the size of the task processed by the user in the pre-encoding stage in the tth time slot is represented as: For each user u n , the calculation time of task representation extracted from the original task n is: wherein is the amount of computation required to extract the task representation, is the computational resource available to extract the task representation for user u n The execution performance of the task and the extraction rate of the task expression are modeled as: When the local computing mode is selected, the computing sub-tasks will be executed locally on the terminal device u n The total execution time locally is the computation time, which is executed in time slot t for task u n The estimated time required is represented as: wherein is the proportion of tasks executed locally, is the computational power size of the local computing task, ψ n is the computational density, D n (t) is the current computing task size; When the service device computation is selected, it is assumed that is the size of the proportion that the service device needs to perform, and u n The sub-tasks performed are offloaded to the service device u d , and executed remotely; u n -u d Uplink transmission: where is the transmission rate, B nd (t) is the bandwidth between the task node u n and the service device u d P nd (t) is the transmission power of the task node u n transmitting the task to the service device u d | h nd (t)| 2 and are the channel coefficient and path loss between the task node u n and the service device u d respectively, σ 2 is the noise power spectral density, u d calculates: where is the computing power size allocated by the service device u d to the task of the task node u n ; u d -u n Downlink transmission: wherein is the downlink transmission rate of the service device u d to the task node u n , is the transmission power of the service device u d to transmit the computation result to the task node u n , is the channel coefficient between the two, γ back is the proportional coefficient of the computation result relative to the original task; ​ u n -u s Uplink transmission: where P is the proportion of tasks executed by the server computing node u ns (t) is the task node u n transmits a subtask to the server computing node u s Transmission power of the subtask, |h ns (t)| 2 and are the channel coefficient and path loss between the two, respectively; u s Computations: where is the server node u s The size of the computing resource provided for the task of the task node u n ; ​ T n The task execution delay corresponding to the computing task is represented as: Assume that there is no overlap between any of the sub-tasks that are satisfied, i.e. satisfied is the proportion of tasks that the task node allocates to all server devices, and is the proportion of tasks that the task node allocates to all server computing nodes; Computational energy consumption to perform a task: The energy consumption of all the computing nodes performing the entire computational task is then where represents the set of all the computing nodes performing the task, κ i is the effective switched capacitance; Communication energy consumption: The non-link task transfer energy consumption of all computing nodes is represented as: Total energy consumption: ​ s.t. ε n ∈(0,1] where ε, A, Y, F are optimization variables, denoted as task expression rate, task association, task partitioning and computing power allocation variables, respectively; the constraint condition is n ∈ (0, 1] represents the average task expression rate range of the task node, is a task association binary variable, represents the execution performance of the task needs to be greater than the threshold represents that the total energy consumption overhead of the computing power node cannot exceed its maximum energy budget represents that there is no overlap between all sub-tasks of the task node, represents the proportion of tasks executed locally by the task node, is the proportion of tasks allocated by the task node to all service devices and, is the proportion of tasks allocated by the task node u n to the service device u d , is the proportion of tasks allocated by the task node to all server computing nodes and, is the proportion of tasks allocated by the task node u n to the server computing node u s , represents the range limit of task partitioning; represents that only when the task is associated with the computing power node can the task be allocated; represents the computing capacity range of the node.

4. The method of claim 3, wherein, ​ The change trend of the task performance and the computational complexity function is analyzed, an expression of the computational complexity and the computational performance and the task expression rate is given, for the additional computational complexity and the task performance function, a fitting algorithm is used for solving, an equivalent analysis is performed on the task pre-coding sub-problem, the task pre-coding sub-problem is converted into a convex optimization problem, the convex optimization problem is solved by using the KKT (Karush-Kuhn-Tucker) condition, and an optimal solution ε of the pre-coding strategy, i.e., the task expression rate, is obtained.

5. The method of claim 4, wherein, The reward function includes an immediate reward and a long-term reward, the immediate reward is a reward for evaluating the result after each task is completed, and involves a mixed reward consistent with the constraint and the optimization goal; the long-term reward balances between the immediate reward and the long-term reward by using a discount factor, and the agent continuously interacts with the environment to maximize the long-term discounted reward; State normalization: state normalization is used to scale the states and place them in the range of [0, 1]; Action mask: all continuous variable constraints of task division, computing power association and computing power allocation are set to [0, 1], and the task division variable is selected from the set (0, 0.01, 0.02, …, 1); The total reward of all computing task nodes is: wherein R n is the reward obtained by the agent after performing an action; The long-term cumulative reward R is calculated as: Wherein γ is a cumulative discount factor.

Images