Demand-differentiated dynamic edge computing network task unloading method and system

Abstract

The invention relates to the technical field of edge computing, and discloses a demand-differentiated dynamic edge computing network task unloading method and system.The method comprises the steps that firstly, a model of a vehicle edge computing system is built, problem description and disassembly are conducted according to an optimization target, and a first problem and a second problem are obtained; and the first problem and the second problem are solved in a staged manner. Thus, through joint optimization of the access decision of the two stages, in the first stage, a vehicle user can make an adaptive roadside unit channel access decision under a dynamic network condition. And in the second stage, the interaction between the vehicle user and the roadside unit is used as a Stackelberg game. Therefore, task scheduling and resource pricing are economically, stably and effectively developed in a balanced manner, and the QoE of the vehicle user and the earnings of the service provider can be maximized while efficient task unloading is met.

Description

Technical Field

[0001] The present invention relates to the field of edge computing technology, and in particular to a method and system for offloading dynamic edge computing network tasks with differentiated requirements. Background Art

[0002] With the rapid development of mobile computing architectures and intelligent sensor technologies, a variety of novel applications have emerged, such as augmented reality (AR), autonomous driving, and environmental awareness. These applications are typically characterized by high computational loads, significant energy consumption, and strict latency constraints, posing substantial challenges to mobile computing platforms. While the processing power of modern in-vehicle embedded systems has significantly improved, they still struggle to meet the exponentially growing computational demands of complex mission scenarios.

[0003] Take autonomous driving as an example. Autonomous vehicles operating on urban roads must perform numerous computationally intensive real-time tasks, such as video recognition for detecting surrounding traffic conditions and online path planning for intelligent driving decisions. Due to complex traffic environments and limited computing power, local processing often results in extended response times, reduced driving efficiency, and even the potential for accidents. Consequently, in-vehicle networks face significant challenges in handling these complex tasks with limited resources.

[0004] In order to alleviate the local computing bottleneck, the academic community has proposed two typical offloading paradigms. One is mobile cloud computing, which has the advantages of powerful data processing capabilities and massive storage capacity. However, the delay caused by long transmission distances and limited bandwidth has brought major challenges to the Internet of Vehicles. Another offloading paradigm is mobile edge computing. Through a distributed architecture, the computing power is dispersed to the edge nodes of the network (such as base stations and roadside units), which can significantly reduce the end-to-end delay while optimizing the traffic load of the backbone network. The vehicle edge computing (VEC) system based on mobile edge computing enables vehicles to offload computing-intensive application data to edge servers, thereby achieving shorter processing delays and lower energy consumption. Collaborative scheduling of heterogeneous resources across on-board terminals and edge servers has become a key technology to improve the efficiency of Internet of Vehicles services.

[0005] However, the high mobility of vehicles, the ever-changing network environment, and the differentiated demands for Quality of Service (QoS) metrics such as latency, energy consumption, and reliability pose significant challenges to resource management. First, in a dynamic network environment, the spatiotemporal imbalance of computing resources is exacerbated. Resources are limited, and resource competition exists on congested roads, making it difficult to obtain optimal task offloading decisions. Furthermore, due to the differentiation of application requirements and the heterogeneity of vehicle computing power, flexibly scheduling vehicle tasks to meet diverse demands is challenging. Finally, the conflict between heterogeneous QoS demands and resource supply hinders the coordinated optimization of user Quality of Experience (QoE) and service provider revenue.

[0006] It can be seen that in the existing technology, edge computing for in-vehicle users is difficult to achieve efficient task offloading while maximizing the QoE of vehicle users and the benefits of service providers. Summary of the Invention

[0007] The present invention provides a dynamic edge computing network task offloading method and system with differentiated requirements to solve the problem in the prior art that edge computing for in-vehicle users is difficult to meet the requirements of efficient task offloading while maximizing the QoE of vehicle users and the benefits of service providers.

[0008] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0009] In a first aspect, the present invention provides a method for offloading dynamic edge computing network tasks with differentiated requirements, comprising:

[0010] S1: Build a model of the vehicle edge computing system, where the model includes a vehicle network model, a vehicle mobility model, a vehicle data transmission model, and a computing model.

[0011] S2: Constructing an optimization target for the vehicle edge computing system;

[0012] S3: Based on the optimization objective, the problem is described and decomposed to obtain a first problem and a second problem. The first problem is a multi-vehicle dynamic access problem, and the second problem is a task scheduling and resource pricing problem.

[0013] S4: solving the first problem, and further solving the second problem based on the solution result;

[0014] S5: Determine the final vehicle dynamic access decision, task scheduling and resource pricing decision based on the results of the first question and the second question.

[0015] In the second aspect, the present application provides a dynamic edge computing network task offloading system with differentiated requirements, comprising a memory, a processor, and a computer program stored on the memory and runnable on the processor, wherein the processor implements the steps of the method described in the first aspect when executing the computer program.

[0016] Beneficial effects:

[0017] The demand-differentiated dynamic edge computing network task offloading method provided by the present invention first constructs a model of the vehicle edge computing system, and describes and decomposes the problem according to the optimization goal to obtain the first problem and the second problem, and solves the first problem and the second problem in a staged manner. In this way, through the two-stage joint optimization of access decisions, in the first stage, the algorithm based on TD 3 enables vehicle users to make adaptive roadside unit channel access decisions under dynamic network conditions. In the second stage, the interaction between vehicle users and roadside units is regarded as a Stackelberg game. This makes task scheduling and resource pricing develop towards an economically stable and effective equilibrium, which can meet efficient task offloading while maximizing the QoE of vehicle users and the benefits of service providers. Combining learning-based access control with economically driven design provides a powerful framework for scalable and adaptive VEC optimization, paving the way for reliable and intelligent transportation systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is one of the flow charts of a demand-differentiated dynamic edge computing network task offloading method according to a preferred embodiment of the present invention;

[0019] Figure 2 This is a second flowchart of a method for offloading dynamic edge computing network tasks with differentiated requirements according to a preferred embodiment of the present invention;

[0020] Figure 3 This is a schematic diagram of the structure of a vehicle edge computing system according to a preferred embodiment of the present invention;

[0021] Figure 4 A diagram illustrating the architecture of a method for solving the first and second problems provided by an embodiment of the present invention;

[0022] Figure 5 TDDA algorithm flow chart provided in an embodiment of the present invention;

[0023] Figure 6 This is a flow chart of the GSP algorithm provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following is a clear and complete description of the technical solutions of the present invention. It should be understood that the embodiments described are only a portion of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0025] Unless otherwise defined, the technical or scientific terms used in the present invention shall have the usual meanings understood by persons of ordinary skill in the field to which the present invention belongs. The words "first", "second" and similar terms used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one" or "a" do not indicate a quantity limitation, but rather indicate the existence of at least one. Words such as "connected" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship also changes accordingly.

[0026] It should be understood that the high mobility of vehicles, the constant changes in the network environment, and the differentiated demands for quality of service (QoS) indicators (such as latency, energy consumption, and reliability) pose significant challenges to resource management. First, in a dynamic network environment, the spatiotemporal imbalance of computing resources is more serious. Resources are limited, there is resource competition in congested roads, and it is difficult to obtain the optimal task offloading decision. In addition, due to the differentiation of application requirements and the heterogeneity of vehicle computing capabilities, there are challenges in flexibly scheduling vehicle tasks for different needs. Finally, the conflict between heterogeneous QoS requirements and resource supply hinders the coordinated optimization of user experience quality (QoE) and service provider revenue. Based on this, the present application provides a task offloading method for a dynamic edge computing network with differentiated demands. To study the task offloading problem in a dynamic edge computing network with differentiated demands. To achieve the purpose of maximizing the QoE of vehicle users and the benefits of service providers while meeting efficient task offloading.

[0027] See Figure 1-Figure 2 , this application provides a demand-differentiated dynamic edge computing network task offloading method, including:

[0028] S1: Build a model of the vehicle edge computing system, where the model includes a vehicle network model, a vehicle mobility model, a vehicle data transmission model, and a computing model.

[0029] S2: Constructing an optimization target for the vehicle edge computing system;

[0030] S3: Based on the optimization objective, the problem is described and decomposed to obtain a first problem and a second problem. The first problem is a multi-vehicle dynamic access problem, and the second problem is a task scheduling and resource pricing problem.

[0031] S4: solving the first problem, and further solving the second problem based on the solution result;

[0032] S5: Determine the final vehicle dynamic access decision, task scheduling and resource pricing decision based on the results of the first question and the second question.

[0033] The aforementioned demand-differentiated dynamic edge computing network task offloading method first constructs a model of the vehicle edge computing system and describes and decomposes the problem according to the optimization objective, resulting in the first and second problems. These problems are then solved in a staged approach. This approach then jointly optimizes access decisions in two stages. In the first stage, a TD3-based algorithm enables vehicle users to make adaptive roadside unit channel access decisions under dynamic network conditions. In the second stage, the interaction between vehicle users and roadside units is formulated as a Stackelberg game. This enables task scheduling and resource pricing to develop towards an economically stable and efficient equilibrium, achieving efficient task offloading while maximizing the quality of experience (QoE) for vehicle users and revenue for service providers. Combining learning-based access control with economically driven design provides a powerful framework for scalable and adaptive VEC optimization, paving the way for reliable and intelligent transportation systems.

[0034] The following describes the steps of the above-mentioned method for offloading tasks in a dynamic edge computing network with differentiated requirements using a specific example:

[0035] 1. Build a model of the vehicle edge computing system, where the model of the vehicle edge computing system includes a vehicle network model, a vehicle mobility model, a vehicle data transmission model, and a computing model.

[0036] 1.1. The vehicle network model is as follows:

[0037] Consider a roadside unit consisting of M roadside units (denoted as R = {r1,…,r M}) and N vehicle users (denoted as V = {v1,…,v N}). Where r i Represents a roadside unit, i takes values ​​of 1...M, v n represents the nth vehicle user, n is 1...N, R represents the set of roadside units, and V represents the set of vehicle users. Figure 3 As shown in Figure 1, each roadside unit is equipped with an edge server to provide computing services within a specific service coverage area. Each vehicle user continuously generates computing tasks when running applications. Typically, vehicle users have limited local computing resources and can offload some tasks to nearby roadside units for processing through wireless channels. The number of wireless channels is K. In this system, time is discretized into a time slot model, represented as T = {0, 1, …, t, …, T-1}, where the length of each time slot is τ. The vehicle's travel speed during the time slot t∈T is In each time slot t∈T, the vehicle user v n ∈V generates a random computation task It is detachable. Indicates the task size in bits. Indicates the task processing density, that is, the number of CPU cycles required to process each bit of task. is the task type parameter, Indicates that the task is delay sensitive, Indicates that the task is energy sensitive.

[0038] Access strategy: is the vehicle user v n Access decision at time slot t. Indicates whether to offload the task. Indicates that the task will not be uninstalled. Indicates that the task is unloaded to the corresponding a-th roadside unit, where the value of a is 1...M. is the vehicle user v n Select the wireless channel. K represents the Kth wireless subchannel, is the access policy for time slot t, i.e., the set of access decisions of all vehicle users.

[0039] Task scheduling strategy: is the vehicle user v n The size of the task unloaded to the roadside unit. The task scheduling strategy is

[0040] Resource allocation strategy: is the vehicle user v n The number of unit computing resources received from the edge server. is the resource cost per unit of computing resources. In addition, the vehicle user v n There is an upper limit on the cost budget of renting computing resources in time slot t Therefore, the resource allocation strategy is:

[0041]

[0042] Where, is the vehicle user v n The number of unit computing resources received from the edge server, n is 1...N, It is the resource cost required for unit computing resources, and n is 1...N.

[0043] 1.2. The vehicle movement model is as follows:

[0044] In the real world, the speed and position of a vehicle will change due to the dynamic conditions of the environment. For dynamic scenarios, a vehicle motion model is established. In time slot t∈T, the vehicle user v n The real-time distance traveled by ∈V on the x and y axes in the two-dimensional coordinate system is

[0045]

[0046] in is the vehicle user v n ∈V The angle between the direction of travel and the x-axis. The initial position in time slot t is When dealing with vehicle computing tasks, represents the change in x-axis displacement from the initial moment to the moment l in time slot t, represents the vehicle's speed during the time slot t∈T, ι represents the time when the vehicle computing task is processed, and the vehicle user v n The position of ∈V changes during this period, and the real-time position coordinates are:

[0047]

[0048] In time slot t∈T, vehicle user v n and roadside units m The real-time distance between them is:

[0049]

[0050] In the formula, (x m ,y m ) represents the coordinates of the roadside unit.

[0051] Each roadside unit r m With coverage When a vehicle user makes an access decision in a single time slot, it needs to ensure that it does not leave the coverage area of ​​the roadside unit to be accessed.

[0052]

[0053] 1.3. The vehicle data transmission model is as follows:

[0054] When vehicle task offloading is implemented, the non-orthogonal multiple access (NOMA) scheme is used to transfer vehicle tasks to roadside units. On this basis, the bandwidth resources of each roadside unit are divided into K sub-channels C = {c1,…,c K}, and each subchannel can be assigned to multiple vehicles.,There is interference when vehicles send data to roadside units through the same,wireless channel. Indicates that in time slot t∈T, the wireless channel c k To the roadside unitm The set of vehicle users that send data. According to the NOMA principle, two or more vehicle users that transmit data at the same time / frequency form a NOMA cluster, i.e. Represents a NOMA cluster. Within the cluster, the signals of vehicle users on the subchannels can be decoded and the interference is partially signaled via successive interference cancellation (SIC). The decoding order depends on the channel gain of the device.

[0055] According to the uplink transmission SIC scheme, the signal with high channel gain is first decoded, and the signal with low channel gain is regarded as interference. The decoded signal is removed from the interference term of other vehicle users with low channel gain. Specifically, at each time instant ι, in time slot t∈T, the wireless channel c K To the roadside unit m The set of vehicle users sending data Sort by channel gain, follow

[0056] in, is the time when vehicle user v n ∈V channel gain. It is v n and r m The straight-line distance between them, δ is the channel fading factor of the communication link. The nth vehicle user can decode the signal of the (n-1)th vehicle user and regard the signal of the (n+1)th vehicle user as noise. Therefore, at time ι, in the wireless channel c k ∈C offloads data to the roadside unit r m The signal-to-interference-and-noise ratio (SINR) is:

[0057]

[0058] Among them, σ 2 is additive white Gaussian noise, p n is the vehicle user v n ∈V task transmission power. Indicates that at time ι, the roadside unit r m The set of covered vehicle users. They are vehicle users v n The intra-cell interference and inter-cell interference are affected. Therefore, the decisions interfere with each other, and different access decisions face different interference. Wireless channel c k The bandwidth is At time ι, vehicle user v n Through channel c k Offload its tasks to roadside units m The data transmission rates are:

[0059]

[0060] 1.4. The calculation model is as follows:

[0061] In the vehicle edge computing system, the tasks generated by vehicle users are considered to be separable. There are two types of task processing, local processing and edge processing. In the time slot t∈T, the vehicle user offloads a certain size of task to the roadside unit for processing according to demand. The amount of offloaded tasks is recorded as Therefore, the latency of local processing is:

[0062]

[0063] Among them, f n is the vehicle user v n The local computing power. The energy consumption generated by local computing is:

[0064]

[0065] where κ is the effective switching capacitance. The vehicle user will calculate part of the task The delay in unloading to a roadside unit for processing is:

[0066]

[0067] Where, At time ι, vehicle user v n Through channel c K Offload its tasks to roadside units m The data transmission rate, is the number of computing resources borrowed by vehicle users from roadside units, f RSU is the unit computing power of the edge server;

[0068] The vehicle user will compute part of the task The energy consumption of unloading to the roadside unit for processing is:

[0069]

[0070] Among them, p n is the vehicle user v n ∈V task transmission power, f RSU is the unit computing power of the edge server, It is the number of units of computing resources borrowed by vehicle users from roadside units.

[0071] 1.5. The utility objectives are as follows:

[0072] This application considers two types of tasks with different QoS requirements: delay-sensitive and energy sensitive The processing performance indicators include:

[0073]

[0074] Where, represents the comprehensive delay of task processing, is the vehicle user v n The access decision at time slot t, is the number of units of computing resources borrowed by vehicle users from roadside units, is the resource cost per unit of computing resources, Indicates the local processing delay time, is part of the vehicle user's computing task Delay time in unloading to roadside units for processing, represents the comprehensive energy consumption of task processing, is part of the vehicle user's computing task Energy consumption offloaded to roadside units for processing.

[0075] Vehicle user v at time slot t n The comprehensive cost is a combination of performance metrics and resource costs, defined in Equation (15).

[0076]

[0077] Where, is the comprehensive cost of task processing, is a task type parameter, θ1 and θ2 weight the performance (i.e., latency and energy consumption) and resource acquisition cost components.

[0078] QoE is used as a user-centric metric for evaluating satisfaction levels. To quantify this relationship, this application uses task processing cost Define a user-centric indicator QOE function for evaluating satisfaction level, where lower cost corresponds to higher QOE. Specifically, vehicle v n The QoE at time t is given by:

[0079]

[0080] Among them, α n Represents vehicle v n QoE growth rate, β n Indicates the baseline processing cost when QoE reaches its median value.

[0081] In the system of this application, the roadside unit sends a nProviding computing resources to earn income. The income of roadside units comes from and allocated computing resources Formally speaking, at time t, a task is processed The income is given by:

[0082]

[0083] 1.6、The problem is described as follows:

[0084] In the scenario of this application, the roadside unit as a resource provider aims to provide (per unit of computing resources) to maximize revenue. At the same time, vehicle users can Price and budget constraints The joint access selection, task scheduling and resource pricing decisions of the network are used to optimize its QoE. This leads to two coupled optimization problems. The first is the QoE optimization problem P1:

[0085]

[0086] Where, represents the user-centric metric QoE used to evaluate the satisfaction level, st. represents the constraint condition, is the vehicle user v n The size of the task unloaded to the roadside unit, Indicates the task size, is the number of units of computing resources borrowed by vehicle users from roadside units, is the resource cost per unit of computing resources, represents the comprehensive delay of task processing, τ is the length of each time slot, It is v n and r m The straight-line distance between is each roadside unit r m With coverage ι indicates the moment;

[0087] The corresponding roadside unit revenue maximization problem P2 is:

[0088]

[0089] Where, represents the revenue of the roadside unit, is the number of units of computing resources borrowed by vehicle users from roadside units, It is the resource cost required for unit computing resources.

[0090] In other words, in this application, P1.2 Together with P2, it constitutes the task scheduling and resource pricing problem. 1.2 is the goal of the vehicle user. P2 is the goal of the roadside unit, which means that the task scheduling and resource pricing problems consider the optimization of two objectives at the same time.

[0091] 2. The dynamic vehicle edge computing network task offloading method with differentiated requirements is as follows:

[0092] The joint optimization of QoE (P1) and revenue (P2) introduces a high-dimensional challenge. Access selection (spatial) and resource pricing (temporal) decisions exhibit strong interdependencies, creating a high-dimensional state-action space. Delay, coverage, and budget constraints impose non-convex restrictions on both vehicle users and roadside units.

[0093] To solve this problem, this paper proposes a two-stage optimization framework TOVEC, which decouples spatiotemporal decision making while maintaining global efficiency. This decomposition combines DRL and game theory to effectively solve the sub-problems: Phase 1: Dynamic access of multiple vehicles (P 1.1 This phase uses DRL to solve real-time access decisions (roadside unit and channel selection) to adapt to dynamic network conditions. Phase 2: Task Scheduling and Resource Pricing (P 1.2 +P2). In this stage, the interaction between vehicle users and roadside units is formulated as a Stackelberg game to jointly optimize the scheduling and pricing strategies.

[0094] 2.1. Multi-vehicle dynamic access is as follows:

[0095] When a vehicle user initiates a task offloading, it must select a roadside unit and a wireless channel. Since task scheduling and resource pricing are not involved at this stage, this application decomposes the access problem into P 1.1 :

[0096]

[0097] Where, Indicates that at time ι, vehicle user v n Through channel c K Offload its tasks to roadside units m Data transfer rate.

[0098] This application will P 1.1 We model it as a Markov Decision Process (MDP) and develop a DRL-based algorithm to learn the optimal access strategy through a series of states, actions, and rewards through environmental interactions. At the beginning of each time slot t, the agent observes the network state s(T), including interference and server availability:

[0099] s(t)={I V (t),I M (t),O M (t),O C (t)} (21)

[0100] Among them, I V (t) is the vehicle user v at time slot t n Transmission interference. M (t)∈{0,1} N×N : Binary interference matrix, if the vehicle user v j v i If interference occurs, O M (t)∈{0,1} N×M : Access availability, if the roadside unit r m For vehicle users n Available, then O M (t) = 1. C (t): the number of roadside units available to each vehicle user, Where m is the roadside unit r m , there are M roadside units. If the roadside unit r m For vehicle users n is optional, then

[0101] Action space: The agent selects an action a consisting of offloading and channel selection acc (t):

[0102] a acc (t) = {a off (t),a cha (t)} (22)

[0103] Among them, a off (t) and a cha (t) represents the roadside unit and channel selection of all vehicle users at time slot t.

[0104] Reward function: The reward function is constructed by the uplink transmission rate obtained by the vehicle user. This reward encourages the minimum transmission rate to be met. strategy while penalizing poor or non-existent decisions:

[0105]

[0106] in, is the vehicle user v during time slot t n The average task transfer rate.

[0107] 2.2 Task scheduling and resource pricing are as follows:

[0108] Given the access decision, this application then solves P 1.2 To determine scheduling and resource pricing under QoE and cost constraints:

[0109]

[0110] Roadside units earn revenue P2 by renting resources to vehicle users. This application models the interaction between resource pricing and task scheduling as a Stackelberg game (RPP) V is the set of vehicle users, and each vehicle user makes a purchase decision R is the set of roadside units, and each roadside unit makes a pricing decision and are the costs to vehicle users and the revenues to roadside units, respectively.

[0111] It is worth emphasizing that in this application, in the game G, there exists a Stackelberg equilibrium, that is, neither the vehicle user nor the roadside unit can improve the QoE and reward by changing the decision.

[0112] In summary, in order to address the dynamic offloading of vehicle edge computing networks with differentiated requirements, this application proposes the TDDA and GSP algorithms to implement the proposed TOVEC framework. These two algorithms jointly optimize vehicle user access, task scheduling, and resource pricing in a dynamic VEC environment. Figure 4 The overall workflow and architecture of the solution process are shown.

[0113] Specifically, the TDDA algorithm is used in this application to solve the access selection problem (P 1.1 ) is modeled as an MDP and DRL is used to capture network dynamics and resource contention, enabling vehicle users to learn the best access strategy to solve the problem. This application designs TDDA as a centralized agent that learns the best access strategy by continuously interacting with the environment. The algorithm flow is shown in the attached Figure 5 As shown in Figure 2, the agent observes the global state, explores the user's access actions, and updates its policy to maximize the long-term reward under uncertainty and fluctuating network conditions. At each time step, the agent observes the current state and chooses an action based on its current policy. Then, it receives the corresponding reward and next state from the environment and sends the experience tuple (s t , a t , r t , s t+1) are stored in a buffer for efficient offline learning. Then, the action and policy networks, as well as their target networks, are iteratively updated. This process enables the agent to converge to a robust access strategy that dynamically adapts to the changing vehicle network topology and resource contention.

[0114] The GSP algorithm combines the task scheduling and resource pricing problems (P 1.2 and P2) are represented as a Stackelberg game between vehicle users and roadside units to solve this problem. Through iterative interaction, both parties converge to the strategy of maximizing vehicle user QoE and roadside unit revenue. The algorithm flow is shown in the attached figure. Figure 6 As shown, the heterogeneous task requirements and the decentralized competition for resources among vehicle users are considered. At each time step, the vehicle user solves the resource request amount and optimal task scheduling decisions The core idea of ​​GSP is to model the competition between vehicle users and roadside units as a repeated game, where each vehicle user is a rational agent whose goal is to maximize their utility within system constraints. A dynamic pricing mechanism is established, allowing roadside units to dynamically adjust resource pricing.

[0115] The present application also provides a demand-differentiated dynamic edge computing network task offloading system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program. The demand-differentiated dynamic edge computing network task offloading system can implement various embodiments of the above method and achieve the same beneficial effects, which will not be described in detail here.

[0116] The above describes in detail the preferred embodiments of the present invention. It should be understood that those skilled in the art can make numerous modifications and variations based on the concepts of the present invention without inventive effort. Therefore, any technical solutions that can be derived by those skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. A dynamic edge computing network task offloading method with differentiated requirements, characterized in that: include: S1: Build a model of the vehicle edge computing system, where the model includes a vehicle network model, a vehicle mobility model, a vehicle data transmission model, and a computing model. S2: Constructing an optimization target for the vehicle edge computing system; S3: Based on the optimization objective, the problem is described and decomposed to obtain a first problem and a second problem. The first problem is a multi-vehicle dynamic access problem, and the second problem is a task scheduling and resource pricing problem. S4: solving the first problem, and further solving the second problem based on the solution result; S5: Determine the final vehicle dynamic access decision, task scheduling and resource pricing decision based on the results of the first question and the second question.

2. The method for offloading dynamic edge computing network tasks with differentiated demand according to claim 1 is characterized in that: The vehicle network model in S1 is as follows: Assume that the vehicle network model includes m roadside units and N vehicle users, which can be expressed as follows: R＝{r1,…,r M}； V={v1,…,v N }; Where r i Represents a roadside unit, i takes values ​​of 1...M, v n represents the nth vehicle user, n is 1...N, R represents the set of roadside units, and V represents the set of vehicle users; Each roadside unit is equipped with an edge server; each vehicle user continuously generates computing tasks when running applications; In the vehicle network model, time is discretized into a time slot model, represented as T = {0, 1, …, t, …, T-1}, where the length of each time slot is τ and the speed of the vehicle during the time slot t∈T is In each time slot t∈T, the vehicle user v n ∈V generates a random computation task in, Indicates the task size in bits; Indicates the number of CPU cycles required to process each bit of task; is the task type parameter, Indicates that the task is delay sensitive, Indicates that the task is energy-sensitive; The vehicle user v n The access decision at time slot t is regarded as the access policy, and the access policy is set as follows: is the vehicle user v n The access decision at time slot t, Indicates whether to offload the task. Indicates that the task will not be uninstalled. Indicates that the task is unloaded to the corresponding a-th roadside unit, where the value of a is 1...M; is the vehicle user v n The selected wireless channel; wherein, K represents the Kth wireless subchannel; the access policy is used to represent the set of access decisions of all vehicle users; The vehicle user v n The task of unloading to the roadside unit is regarded as a task scheduling strategy. The task scheduling strategy is set as follows: Where, is the vehicle user v n The size of the task unloaded to the roadside unit, where n is 1...N; The vehicle user v n The case of receiving resources from the edge server is considered as the resource allocation strategy. The resource allocation strategy is set as follows: Where, is the vehicle user v n The number of unit computing resources received from the edge server, n is 1...N, It is the resource cost required for unit computing resources, and the value of n is 1...N.

3. The method for offloading dynamic edge computing network tasks with differentiated demand according to claim 2 is characterized in that: The vehicle movement model in S1 is as follows: Assume that in time slot t∈T, vehicle user v n The real-time distance ∈V travels on the x and y axes in a two-dimensional coordinate system is as follows: in, is the vehicle user v n The angle between the traveling direction of ∈V and the x-axis direction, the initial position in time slot t is represents the change in x-axis displacement from the initial moment to the moment l in time slot t, represents the vehicle's speed during the time slot t∈T, ι represents the time when the vehicle computing task is processed, and the vehicle user v n The position of ∈V changes during this period, and the real-time position coordinates are: In time slot t∈T, vehicle user v n and roadside units m The real-time distance between them is: In the formula, (x m ,y m ) represents the coordinates of the roadside unit; Set each roadside unit r m With coverage When a vehicle user makes an access decision in a single time slot, it needs to ensure that it does not leave the coverage of the roadside unit to be accessed, and the following relationship is satisfied:

4. The method for offloading dynamic edge computing network tasks with differentiated demand according to claim 1, characterized in that: The vehicle data transmission model in S1 is as follows: When vehicle task offloading is implemented, a non-orthogonal multiple access scheme is used to transfer the vehicle task to the roadside unit; and the bandwidth resources of each roadside unit are divided into K sub-channels C = {c1,…,c K }, where c K represents the Kth subchannel; each subchannel can be assigned to multiple vehicles; Assume that at each time instant ι, in time slot t∈T, the wireless channel c K To the roadside unit m The set of vehicle users sending data The order of channel gain is as follows: in, is the time when vehicle user v n ∈V channel gain, n is It is v n and r m The straight-line distance between them, δ is the channel fading factor of the communication link. The nth vehicle user can decode the signal of the (n-1)th vehicle user and regard the signal of the (n+1)th vehicle user as noise. Therefore, at time ι, in the wireless channel c k ∈C offloads data to the roadside unit r m The signal-to-interference-noise ratio is: Among them, σ 2 is additive white Gaussian noise, p n is the vehicle user v n ∈V task transmission power, Indicates that at time ι, the roadside unit r m The set of covered vehicle users, and They are vehicle users v n Intra-cell interference and inter-cell interference; Wireless channel c k The bandwidth is At time ι, vehicle user v n Through channel c k Offload its tasks to roadside units m The data transmission rates are:

5. The method for offloading dynamic edge computing network tasks with differentiated demand according to claim 2, characterized in that: The calculation model in S1 is as follows: Assuming that in the time slot t∈T, the vehicle user unloads a certain size of task to the roadside unit according to demand, and the unloaded task amount is recorded as The delay of local processing satisfies the following relationship: in, represents the local processing delay time, f n is the vehicle user v n The local computing power, the energy consumption generated by local computing is: Where κ is the effective switching capacitance, the vehicle user will calculate part of the task The delay in unloading to a roadside unit for processing is: Where, At time ι, vehicle user v n Through channel c K Offload its tasks to roadside units m The data transmission rate, is the number of computing resources borrowed by vehicle users from roadside units, f RSU is the unit computing power of the edge server; The vehicle user will compute part of the task The energy consumption of unloading to the roadside unit for processing is: Among them, p n is the vehicle user v n ∈V task transmission power.

6. The method for offloading dynamic edge computing network tasks with differentiated demand according to claim 1, characterized in that: The S2 includes: The task types include delay-sensitive and energy-sensitive, and the processing performance indicators include: Where, represents the comprehensive delay of task processing, is the vehicle user v n The access decision at time slot t, is the number of units of computing resources borrowed by vehicle users from roadside units, is the resource cost per unit of computing resources, Indicates the local processing delay time, is part of the vehicle user's computing task Delay time in unloading to roadside units for processing, represents the comprehensive energy consumption of task processing, is the energy consumption of local processing tasks, is part of the vehicle user's computing task Energy consumption when offloading to roadside units for processing; Vehicle user v at time slot t n The comprehensive cost of task processing is a combination of performance metrics and resource costs, defined as follows: Where, is the comprehensive cost of task processing, is the task type parameter, θ1 represents the weighted component of the performance metric, and θ2 represents the weighted component of the resource acquisition cost; Comprehensive cost based on task processing Define a user-centric indicator QOE function for evaluating satisfaction level, where lower cost corresponds to higher QOE, and vehicle v n The QoE at time t satisfies the following: in, represents the QOE function, α n Indicates vehicle v n QoE growth rate, β n It represents the baseline processing cost when QoE reaches its median value; Processing a task at time t The income is as follows: Optimization goals are set based on revenue and QoE as follows: Roadside units act as resource providers through dynamic pricing To maximize revenue, vehicle users jointly optimize access selection, task scheduling, and resource pricing decisions based on task type, price, and budget constraints to maximize QoE.

7. The method for offloading dynamic edge computing network tasks with differentiated demand according to claim 1, characterized in that: The S3 includes: According to the optimization goal, the optimization problem P1 of user-centric QoE metric for evaluating satisfaction level is obtained as follows: Where, represents the user-centric QOE function for evaluating satisfaction levels, st. represents the constraints, is the vehicle user v n The size of the task unloaded to the roadside unit, Indicates the task size, is the number of units of computing resources borrowed by vehicle users from roadside units, is the resource cost per unit of computing resources, is the vehicle user v in the current time slot t n The resource purchase budget cap is represents the comprehensive delay of task processing, τ is the length of each time slot, It is v n and r m The straight-line distance between is each roadside unit r m With coverage ι indicates the moment; According to the optimization objective, the target P2 of the roadside unit is obtained, where the target P2 of the roadside unit is the task scheduling and resource pricing problem of the roadside unit, as follows: Where, represents the revenue of the roadside unit, is the number of units of computing resources borrowed by vehicle users from roadside units, It is the resource cost required for unit computing resources; Split the problem P1 into the multi-vehicle dynamic access problem P 1.1 and the vehicle user's target P 1.2 , where the vehicle user’s target P 1.2 Task scheduling and resource pricing problems for vehicle users; The problem of dynamically connecting multiple vehicles to P 1.1 As the first question in the first stage, P 1.2 and question P2 as the second question of the second stage.

8. The method for offloading dynamic edge computing network tasks with differentiated demand according to claim 7 is characterized in that: The S4 includes: Set P 1.1 as follows: Where, Indicates that at time ι, vehicle user v n Through channel c K Offload its tasks to roadside units m Data transfer rate; P 1.1 Modeled as an MDP, the optimal access policy is learned using the deep reinforcement learning algorithm DRL. At the beginning of each time slot t, the agent observes the network state s(t), including interference and server availability: s(t)={I V (t),I M (t),O M (t),O C (t)}; Among them, I V (t) is the vehicle user v at time slot t n Transmission interference, I M (t)∈{0,1} N×N is a binary interference matrix, if the vehicle user v j v i If interference occurs, is access availability; if the roadside unit r m For vehicle users n Available, then O M (t) = 1, O C (t) the number of roadside units available to each vehicle user; Set action space: The agent selects an action consisting of offloading and channel selection a acc (t) are as follows: a acc (t)={a off (t),a cha (t)}; Among them, a off (t) and a cha (t) represents the roadside unit and channel selection of all vehicle users at time slot t; Set reward function: Construct a reward function based on the uplink transmission rate obtained by the vehicle user, which encourages the minimum transmission rate to be met strategy while penalizing poor or non-existent decisions: in, is the vehicle user v during time slot t n The average task transfer rate; Given the access decision, solve P 1.2 To determine the scheduling and resource pricing under QoE and cost constraints, as follows: The roadside unit obtains income P2 by renting resources to vehicle users at a certain price. The interaction between resource pricing and task scheduling is modeled as a Stackelberg game (RPP) G, where V is the set of vehicle users, and each vehicle user makes a purchase decision R is the set of roadside units, and each roadside unit makes a pricing decision and are the costs to vehicle users and the revenues to roadside units, respectively; In the game G, there exists a Stackelberg equilibrium such that neither the vehicle user nor the roadside unit can improve QoE and returns by changing their decisions.

9. A dynamic edge computing network task offloading system with differentiated requirements, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images