An edge computing resource allocation method for QoE

Abstract

This invention discloses a QoE-oriented edge computing resource allocation method, comprising: constructing a QoE-oriented resource allocation model based on different QoE requirement characteristics, the resource allocation model including an end-user resource allocation scheme and an edge server resource allocation scheme; the end-user resource allocation adopts an offload decision mechanism, whereby users obtain the optimal computing resource allocation according to their QoE requirements and process tasks when the QoE requirements can be met; otherwise, the tasks are offloaded to the edge server with the optimal transmission power for communication scheduling; the edge server resource allocation employs multi-objective reinforcement learning to jointly schedule the communication and computing resources of the edge server to meet the different QoE requirements of end users. The method proposed in this invention performs reasonable resource allocation to meet the different QoE requirements of edge-assisted IoT applications, improves the QoE value, and enables end users and the edge computing system to jointly schedule communication and computing resources, thereby maximizing the overall QoE among multiple IoT users.

Description

Technical Field

[0001] This invention relates to the field of resource allocation strategy technology, and in particular to an edge computing resource allocation method oriented towards QoE. Background Technology

[0002] With the proliferation of the Internet of Things (IoT) and the widespread penetration of wireless networks, the demand for data communication and computing has surged, necessitating the emerging edge computing paradigm. By moving services and functions from the cloud closer to the user, edge computing can provide powerful communication, storage, networking, and connectivity capabilities. Resource allocation in edge computing is crucial to the success of edge computing systems.

[0003] In reality, edge computing can be affected by various factors, such as available bandwidth, edge capabilities and availability, user device capabilities, and the quality of radio and backhaul connections. Therefore, computing and communication resources on edge computing servers should be shared among compute offloaded components. Furthermore, different applications pursue different QoE requirements during task processing; for example, minimizing computational latency for driving safety, or minimizing energy consumption to extend battery life. Therefore, rationally allocating resources to meet the diverse QoE requirements of edge-assisted IoT applications is one of the urgent problems to be solved. Summary of the Invention

[0004] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by this invention is: edge computing is affected by various factors, which affect computing latency, and the problem of extending battery life and minimizing energy consumption.

[0007] To solve the above technical problems, the present invention provides the following technical solution: collecting user perception feature data, constructing a QoE-oriented resource allocation model based on different QoE demand features, wherein the resource allocation model has a two-stage resource allocation scheme, namely, resource allocation for end users and resource allocation for edge servers;

[0008] The terminal user resource allocation adopts an effective and low-complexity offloading decision mechanism. Each user can obtain the best computing resource allocation according to their own QoE requirements and process tasks if the QoE requirements can be met. Otherwise, the task will be offloaded to the edge server with the best transmission power of communication scheduling.

[0009] The resource allocation of the edge servers adopts multi-objective reinforcement learning to jointly schedule the communication and computing resources of the edge servers in order to meet the different QoE needs of end users.

[0010] As a preferred embodiment of the QoE-oriented edge computing resource allocation method of the present invention, the resource allocation model includes a task model, a local computing model, an edge computing model, and a QoE evaluation model.

[0011] As a preferred embodiment of the QoE-oriented edge computing resource allocation method described in this invention, the task model specifically includes:

[0012] The workload of the end user is modeled as a Poisson process, with the rate set as λ. n During time period T, end user n has A computational task queue with a QoE requirement Q n Matrix Q n Represents task queue A n Constraints and importance of each QoE factor in the table;

[0013] Each computational task is described as a triple, which includes the task's arrival time t. n,i The amount of data to be processed and calculation workload All tasks waiting in the queue are generated by applications with different QoE requirements.

[0014] As a preferred embodiment of the QoE-oriented edge computing resource allocation method described in this invention, the local computing model specifically includes:

[0015] When computation task A n,i When executed locally, each end user is assigned to process computation task A. n,i The processing power is f n,i The total latency of local computation consists of waiting time and computation time.

[0016] As a preferred embodiment of the QoE-oriented edge computing resource allocation method described in this invention, the edge computing model specifically includes:

[0017] When end user n is unable to handle the computational task A that meets its QoE requirements n,i At that time, A n,i It will be added to the task queue of the edge computing system for further processing;

[0018] The processing in an edge computing system mainly consists of communication and computation components;

[0019] The communication component includes edge computing servers receiving computing tasks via wireless communication links;

[0020] The computing component includes edge computing servers that process computing tasks.

[0021] As a preferred embodiment of the QoE-oriented edge computing resource allocation method described in this invention, the QoE evaluation model specifically includes:

[0022] Two types of QoE factors are defined: the proportional factor QoE-PF and the inverse proportional factor QoE-IPF.

[0023] For QoE-PF, from the lower bound to the upper bound, QoE maintains a steady increase as the factor increases; when the factor reaches the upper bound, close to the highest QoE level, QoE remains almost unchanged.

[0024] For QoE-IPF, a higher value generally reduces the QoE level, and as the factor increases, QoE gradually decreases from the lower limit to the upper limit.

[0025] Choose a linear function q k As a QoE function

[0026]

[0027] Where v represents the QoE performance of each factor, v low and v up These represent the lower and upper limits of the QoE requirement, respectively;

[0028] Computation Task A n,i The QoE performance is in, This means that 0 ≤ α K ≤1 and The importance of the QoE factor The q-function values ​​represent different factors.

[0029] As a preferred embodiment of the QoE-oriented edge computing resource allocation method described in this invention, it includes:

[0030] After the application task is generated, end user n selects the head of the task queue to compute task A. n,i To obtain the local optimal computing resources and corresponding optimal QoE performance of end user n;

[0031] If the optimal QoE performance can meet the user's needs, then the end user n will use the optimal computing resources locally to process his / her service requests.

[0032] If the optimal QoE performance obtained through optimal computing resources cannot meet the user's QoE requirements, then the communication is offloaded to the edge server at the optimal transmission power of the communication schedule.

[0033] When an edge computing system receives application tasks from end users with different QoE requirements, it needs to combine communication and computing resources to meet the QoE requirements of the end users.

[0034] As a preferred embodiment of the QoE-oriented edge computing resource allocation method described in this invention, the specific operations of the terminal user resource allocation include:

[0035] An effective and low-complexity offloading decision mechanism is adopted, which allows each user to obtain the optimal allocation of computing resources according to their own QoE requirements and process tasks while meeting the QoE requirements.

[0036] The unloading decision mechanism utilizes the optimal CPU cycle frequency. Calculate the optimal QoE;

[0037] If even the optimal QoE cannot meet the user's QoE requirements, then the optimal transmission power for communication scheduling will be used. Offload to an edge server;

[0038] Given task information and its QoE requirements, it can be represented as follows:

[0039]

[0040] in, This represents the lower limit of the optimal CPU cycle frequency. α represents the lower limit of the optimal CPU cycle frequency. n Indicating the importance of the QoE delay factor, 1-α n This indicates the importance of the QoE energy consumption factor;

[0041]

[0042] in, k represents the optimal CPU cycle frequency without considering upper and lower bound constraints. n It is a coefficient that reflects the relationship between end-user computing power and energy consumption. This represents the maximum energy consumption constraint for the task. This represents the maximum delay constraint for the i-th computation task of the n-th user.

[0043] As a preferred embodiment of the QoE-oriented edge computing resource allocation method described in this invention, it includes:

[0044] Optimal transmission power This can be represented by given task information and its QoE requirements:

[0045]

[0046] in, This represents the lower limit of the optimal transmission power. Indicates the upper limit of optimal transmission power. This represents the optimal transmission power without considering upper and lower limit constraints.

[0047] in satisfy:

[0048]

[0049] Where h represents the channel fading coefficient, ω0 represents the Gaussian white noise power, and d -θ d represents path loss. θ The value represents the reciprocal of path loss, where d represents the distance from the end user to the edge server, θ represents the path loss exponent, and E represents the path loss exponent. max Indicates energy consumption limit, T max Indicates a delay limit.

[0050] As a preferred embodiment of the QoE-oriented edge computing resource allocation method described in this invention, the resource allocation operation of the edge server specifically includes:

[0051] Multi-objective reinforcement learning is used to jointly schedule edge server communication and computing resources to meet the different QoE needs of end users;

[0052] The multi-objective reinforcement learning includes employing the PPO (Policy Nearest Optimization) algorithm to meet the different QoE requirements of end users and maximize QoE performance.

[0053] The resource scheduling problem of edge servers can be described as an MDP (Markov Decision Process), where the set of end users is regarded as the environment, and the edge computing system plays the role of an agent. The agent continuously executes decisions and interacts with the environment.

[0054] The environment of the MDP model contains several key elements, including state, action, and reward. The application's state information and available resources are used as state; action refers to the agent scheduling communication and computing resources according to the current state; the environment responds to actions and enters a new state with a reward signal, the design of which conforms to the optimization objective.

[0055] For each time step n, the agent observes the environmental state s in the state space. n and follow strategy π(a n|s n Select action a from action space A. n The strategy π is in the observed state s n Take action a at that time n The probability; then, the environment transitions to the next state s. n+1 And according to P(s) n+1 |s n ,a n ) and reward function R(s n ,a n Send a reward signal to the agent. n Unless the agent observes a final state, this process will continue indefinitely.

[0056] Establish a neural network structure to store the data generated by the agent's interaction with the environment in an experience pool;

[0057] The goal of optimization is to meet the QoE requirements of end users and maximize the QoE value. Based on the optimization goal, a batch of data is continuously selected from the experience pool to train the neural network, thereby optimizing the neural network parameters.

[0058] When an edge computing system receives application tasks from end users with different QoE requirements, it can obtain the optimal allocation strategy through an optimized neural network. The edge computing system then combines communication and computing resources according to the optimal strategy to meet the QoE requirements of the end users and obtain the optimal QoE value.

[0059] The beneficial effects of this invention are: the method proposed in this invention solves the problem of reasonable resource allocation to meet the different QoE requirements of edge-assisted IoT applications, improves the QoE value, and enables end users and edge computing systems to jointly schedule communication and computing resources, thereby maximizing the overall QoE among multiple IoT users. Attached Figure Description

[0060] 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. Wherein:

[0061] Figure 1 This is a basic flowchart illustrating a QoE-oriented edge computing resource allocation method according to an embodiment of the present invention.

[0062] Figure 2 This is a schematic diagram of the structure of a QoE-oriented resource allocation model for an edge computing resource allocation method provided in an embodiment of the present invention;

[0063] Figure 3 This is a schematic diagram of the structure of a two-stage resource allocation method for QoE-oriented edge computing provided in an embodiment of the present invention;

[0064] Figure 4 This is an experimental simulation diagram of an edge computing resource allocation method for QoE provided in one embodiment of the present invention. Detailed Implementation

[0065] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0066] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0067] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0068] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0069] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0070] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0071] Example 1

[0072] Reference Figures 1-3 As an embodiment of the present invention, a method for allocating edge computing resources for QoE is provided, comprising:

[0073] S1: Collect user perception feature data and construct a QoE-oriented resource allocation model based on different QoE demand features. The resource allocation model has a two-stage resource allocation scheme, namely, resource allocation for end users and resource allocation for edge servers.

[0074] S2: End-user resource allocation employs an efficient and low-complexity offloading decision mechanism. Each user can obtain the optimal allocation of computing resources based on their own QoE requirements and process tasks while meeting QoE requirements; otherwise, tasks will be offloaded to edge servers at the optimal transmission power for communication scheduling. It should be noted that:

[0075] Resource allocation models include task models, local computing models, edge computing models, and QoE evaluation models.

[0076] The task model specifically includes,

[0077] The workload of the end user is modeled as a Poisson process, with the rate set as λ. n During time period T, end user n has A computational task queue with a QoE requirement Q n Matrix Q n Represents task queue A n Constraints and importance of each QoE factor in the table;

[0078] Each computational task is described as a triple, which includes the task's arrival time t. n,i The amount of data to be processed and calculation workload All tasks waiting in the queue are generated by applications with different QoE requirements.

[0079] The local computing model specifically includes,

[0080] When computation task A n,iWhen executed locally, each end user is assigned to process computation task A. n,i The processing power is f n,i The total latency of local computation consists of waiting time and computation time.

[0081] The edge computing model specifically includes,

[0082] When end user n is unable to handle the computational task A that meets its QoE requirements n,i At that time, A n,i It will be added to the task queue of the edge computing system for further processing;

[0083] The processing in an edge computing system mainly consists of communication and computation components;

[0084] The communication component includes edge computing servers receiving computing tasks via wireless communication links;

[0085] The computing component includes edge computing servers that process computing tasks.

[0086] The QoE evaluation model specifically includes:

[0087] Two types of QoE factors are defined: the proportional factor QoE-PF and the inverse proportional factor QoE-IPF.

[0088] For QoE-PF, from the lower bound to the upper bound, QoE maintains a steady increase as the factor increases; when the factor reaches the upper bound, close to the highest QoE level, QoE remains almost unchanged.

[0089] For QoE-IPF, a higher value generally reduces the QoE level, and as the factor increases, QoE gradually decreases from the lower limit to the upper limit.

[0090] Choose a linear function q k As a QoE function

[0091]

[0092] Where v represents the QoE performance of each factor, v low and v up These represent the lower and upper limits of the QoE requirement, respectively;

[0093] Computation Task A n,i The QoE performance is in, This means that 0 ≤ α K ≤1 and The importance of the QoE factor The q-function values ​​represent different factors.

[0094] include,

[0095] After the application task is generated, end user n selects the head of the task queue to compute task A. n,i To obtain the local optimal computing resources and corresponding optimal QoE performance of end user n;

[0096] If the optimal QoE performance can meet the user's needs, then the end user n will use the optimal computing resources locally to process his / her service requests.

[0097] If the optimal QoE performance obtained through optimal computing resources cannot meet the user's QoE requirements, then the communication is offloaded to the edge server at the optimal transmission power of the communication schedule.

[0098] When an edge computing system receives application tasks from end users with different QoE requirements, it needs to combine communication and computing resources to meet the QoE requirements of the end users.

[0099] S3: Edge server resource allocation employs multi-objective reinforcement learning to jointly schedule edge server communication and computing resources to meet the diverse QoE needs of end users. It should be noted that:

[0100] The specific operations for allocating end-user resources include,

[0101] An effective and low-complexity offloading decision mechanism is adopted, which allows each user to obtain the optimal allocation of computing resources according to their own QoE requirements and process tasks while meeting the QoE requirements.

[0102] The unloading decision mechanism utilizes the optimal CPU cycle frequency Calculate the optimal QoE;

[0103] If even the optimal QoE cannot meet the user's QoE requirements, then the optimal transmission power for communication scheduling will be used. Offload to an edge server;

[0104] Given task information and its QoE requirements, it can be represented as follows:

[0105]

[0106] in, This represents the lower limit of the optimal CPU cycle frequency. α represents the lower limit of the optimal CPU cycle frequency. n Indicating the importance of the QoE delay factor, 1-α n This indicates the importance of the QoE energy consumption factor;

[0107]

[0108] in, k represents the optimal CPU cycle frequency without considering upper and lower bound constraints. n It is a coefficient that reflects the relationship between end-user computing power and energy consumption. This represents the maximum energy consumption constraint for the task. This represents the maximum delay constraint for the i-th computation task of the n-th user.

[0109] include,

[0110] Optimal transmission power This can be represented by given task information and its QoE requirements:

[0111]

[0112] in, This represents the lower limit of the optimal transmission power. Indicates the upper limit of optimal transmission power. This represents the optimal transmission power without considering upper and lower limit constraints.

[0113] in satisfy:

[0114]

[0115] Where h represents the channel fading coefficient, ω0 represents the Gaussian white noise power, and d -θ d represents path loss. θ The value represents the reciprocal of path loss, where d represents the distance from the end user to the edge server, θ represents the path loss exponent, and E represents the path loss exponent. max Indicates energy consumption limit, T max Indicates a delay limit.

[0116] The resource allocation operations for edge servers specifically include,

[0117] Multi-objective reinforcement learning is used to jointly schedule edge server communication and computing resources to meet the different QoE needs of end users;

[0118] Multi-objective reinforcement learning includes employing the PPO (Policy Nearest Optimization) algorithm to meet the different QoE requirements of end users and maximize QoE performance;

[0119] The resource scheduling problem of edge servers can be described as an MDP (Markov Decision Process), where the set of end users is regarded as the environment, and the edge computing system plays the role of an agent. The agent continuously executes decisions and interacts with the environment.

[0120] The environment of the MDP model contains several key elements, including state, action, and reward. The application's state information and available resources are used as state; action refers to the agent scheduling communication and computing resources according to the current state; the environment responds to actions and enters a new state with a reward signal, the design of which conforms to the optimization objective.

[0121] For each time step n, the agent observes the environmental state s in the state space. n and follow strategy π(a n |s n Select action a from action space A. n The strategy π is in the observed state s n Take action a at that time n The probability; then, the environment transitions to the next state s. n+1 And according to P(s) n+1 |s n ,a n ) and reward function R(s n ,a n Send a reward signal to the agent. n Unless the agent observes a final state, this process will continue indefinitely.

[0122] Establish a neural network structure to store the data generated by the agent's interaction with the environment in an experience pool;

[0123] The goal of optimization is to meet the QoE requirements of end users and maximize the QoE value. Based on the optimization goal, a batch of data is continuously selected from the experience pool to train the neural network, thereby optimizing the neural network parameters.

[0124] When an edge computing system receives application tasks from end users with different QoE requirements, it can obtain the optimal allocation strategy through an optimized neural network. The edge computing system then combines communication and computing resources according to the optimal strategy to meet the QoE requirements of the end users and obtain the optimal QoE value.

[0125] The specific process of the two-phase resource allocation scheme for QoE is as follows:

[0126] A1: Initialize computing and communication resources for end users and edge computing systems.

[0127] A2: Initialize the tasks generated by each end user.

[0128] A3: / *Phase One – End-User Allocation Scheme* /

[0129] A4: for n=1,2,...,N do

[0130] A5: End user n selects the head of the task queue to calculate task A. n,i Obtain its local optimal CPU cycle frequency and the corresponding optimal QoE performance

[0131] A6: if optimal QoE performance is less than user QoE requirements, then

[0132] A7: With optimal transmission power A n,i Unloaded to the task queue of the edge server

[0133] A8: end if

[0134] A9: endfor

[0135] A10: / *Phase Two – Edge Computing Allocation Scheme* /

[0136] A11: Sort all unloaded tasks in the task queue by arrival time.

[0137] A12: for epoch=1,2,...,K do

[0138] A13: Allocate resources to tasks in the task queue and interact with the environment.

[0139] A14: Update the edge server allocation strategy based on environmental feedback.

[0140] A15: end for

[0141] The method proposed in this invention addresses the issue of reasonable resource allocation to meet the different QoE requirements of edge-assisted IoT applications, thereby improving the QoE value. It enables end users and edge computing systems to jointly schedule communication and computing resources, maximizing the overall QoE among multiple IoT users.

[0142] Example 2

[0143] Reference Figure 4 This is the second embodiment of the present invention. Unlike the first embodiment, this embodiment provides a verification test of an edge computing resource allocation method for QoE. To verify and explain the technical effects of the method, this embodiment compares the traditional technical solution with the method of the present invention, and compares the test results with scientific demonstration methods to verify the real effect of the method.

[0144] To verify the effectiveness of this method, simulation experiments were conducted. The algorithm and network architecture were implemented using TensorFlow, and the plots were drawn using Matplotlib. The performance of the two-stage resource allocation method (TS) was evaluated by comparing it with three representative benchmark methods: a delay-first algorithm (LS), an energy-first algorithm (ES), and a randomized method (Rand). The results are shown in the figure below. This figure shows that for different QoE factors, the two-stage resource allocation method in this paper achieves a higher QoE value than the other three methods, thus verifying the superiority of this method.

[0145] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for QoE-oriented edge computing resource allocation, characterized in that, include: Collect user perception feature data and construct a QoE-oriented resource allocation model based on different QoE demand features. The resource allocation model has a two-stage resource allocation scheme, namely, resource allocation for end users and resource allocation for edge servers. The end-user resource allocation employs an efficient and low-complexity offloading decision mechanism. Each user can obtain the optimal computing resource allocation based on their own QoE requirements and process tasks while meeting the QoE requirements; otherwise, the task will be offloaded to the edge server at the optimal transmission power for communication scheduling. The specific operations of the end-user resource allocation include... An effective and low-complexity offloading decision mechanism is adopted, which allows each user to obtain the optimal allocation of computing resources according to their own QoE requirements and process tasks while meeting the QoE requirements. The offloading decision mechanism utilizes an optimal CPU cycle frequency , calculates an optimal QoE; If the optimal QoE cannot meet the user QoE requirement, the best transmission power of communication scheduling is taken offloaded to an edge server; By giving the task information and its QoE requirement is expressed as: ； wherein, represents a lower limit of an optimal CPU cycle frequency, represents an upper limit of an optimal CPU cycle frequency, represents the importance of a QoE delay factor, represents the importance of a QoE energy consumption factor; ； in, This represents the optimal CPU cycle frequency without considering upper and lower bound constraints. It is a coefficient that reflects the relationship between end-user computing power and energy consumption. This represents the maximum energy consumption constraint for the task. Indicates the first n The user i Maximum latency constraint for each computational task; Optimal transmission power This can be represented by given task information and its QoE requirements: ； in, This indicates the lower limit of the optimal transmission power. Indicates the upper limit of optimal transmission power. This represents the optimal transmission power without considering upper and lower limit constraints. in satisfy: ； in, h Represents the channel fading coefficient. This represents the power of Gaussian white noise. Indicates path loss. This represents the reciprocal of the path loss. d This represents the distance from the end user to the edge server. This represents the path loss index. Indicates energy consumption limit, Indicates delay limit; The resource allocation of the edge servers employs multi-objective reinforcement learning to jointly schedule the communication and computing resources of the edge servers to meet the different QoE needs of end users. The specific operations of the edge server resource allocation include... Multi-objective reinforcement learning is used to jointly schedule edge server communication and computing resources to meet the different QoE needs of end users; The multi-objective reinforcement learning includes employing the PPO algorithm to meet the different QoE needs of end users and maximize QoE performance; The resource scheduling problem of edge servers is described as MDP, where the set of end users is regarded as the environment, and the edge computing system plays the role of agent, which continuously executes decisions and interacts with the environment; The environment of the MDP model contains several key elements, including state, action, and reward. The application's state information and available resources are used as state; action refers to the agent scheduling communication and computing resources according to the current state; the environment responds to actions and enters a new state with a reward signal, the design of which conforms to the optimization objective. For each time step n The agent observes the environment state in the state space. and follow the strategy Select an action from action space A. Strategy π is in the observed state Take action at the time The probability; then, the environment transitions to the next state. and respectively according to and reward function Send reward signals to agents Unless the agent observes a final state, this process will continue indefinitely. Establish a neural network structure to store the data generated by the agent's interaction with the environment in an experience pool; The goal of optimization is to meet the QoE requirements of end users and maximize the QoE value. Based on the optimization goal, a batch of data is continuously selected from the experience pool to train the neural network, thereby optimizing the neural network parameters. When an edge computing system receives application tasks from end users with different QoE requirements, it can obtain the optimal allocation strategy through an optimized neural network. The edge computing system then combines communication and computing resources according to the optimal strategy to meet the QoE requirements of the end users and obtain the optimal QoE value.

2. The edge computing resource allocation method for QoE as described in claim 1, characterized in that: The resource allocation model includes a task model, a local computing model, an edge computing model, and a QoE evaluation model.

3. The edge computing resource allocation method for QoE as described in claim 2, characterized in that: The task model specifically includes, The end-user workload is modeled as a Poisson process, with the rate set as... During time period T, end users n have A computational task queue with QoE requirements. ,matrix Represents the task queue Constraints and importance of each QoE factor in the table; Each computational task is described as a triple, which includes the task's arrival time. The amount of data to be processed and calculation workload All tasks waiting in the queue are generated by applications with different QoE requirements.

4. The edge computing resource allocation method for QoE as described in claim 3, characterized in that: The local computing model specifically includes, When computing tasks When executed locally, each end user is assigned a computational task. Its processing power is The total latency of local computation consists of waiting time and computation time.

5. The edge computing resource allocation method for QoE as described in claim 4, characterized in that: The edge computing model specifically includes, When end users n Unable to handle computational tasks under its QoE requirements. hour, It will be added to the task queue of the edge computing system for further processing; The processing in an edge computing system mainly consists of communication and computation components; The communication component includes edge computing servers receiving computing tasks via wireless communication links; The computing component includes edge computing servers that process computing tasks.

6. The edge computing resource allocation method for QoE as described in any one of claims 2 to 5, characterized in that: The QoE evaluation model specifically includes: Two types of QoE factors are defined: the proportional factor QoE-PF and the inverse proportional factor QoE-IPF. For QoE-PF, from the lower bound to the upper bound, QoE maintains a steady increase as the factor increases; when the factor reaches the upper bound, close to the highest QoE level, QoE remains almost unchanged. For QoE-IPF, a higher value generally reduces the QoE level, and as the factor increases, QoE gradually decreases from the lower limit to the upper limit. Choose a linear function q k As a QoE function ； in, v This represents the QoE performance of each factor. and These represent the lower and upper limits of the QoE requirement, respectively; Computational tasks The QoE performance is ,in, Indicates satisfaction and The importance of the QoE factor , Indicating different factors q function value, .

7. The edge computing resource allocation method for QoE as described in claim 6, characterized in that: include, After the application task is generated, the end user n Select the head of the task queue to calculate the task. Acquire end users n The local optimal computing resources and corresponding optimal QoE performance; If optimal QoE performance can meet the user's needs, then the end user n Use the optimal computing resources locally to process your service requests; If the optimal QoE performance obtained through optimal computing resources cannot meet the user's QoE requirements, then the communication is offloaded to the edge server at the optimal transmission power of the communication schedule. When an edge computing system receives application tasks from end users with different QoE requirements, it needs to combine communication and computing resources to meet the QoE requirements of the end users.

Images