Edge computing task unloading method based on graph theory

Abstract

The invention relates to an edge computing task unloading method based on a graph theory. The method comprises the following steps: S1, initializing parameters of a system; s2, calculating task time delay and energy consumption; s3, task classification based on time delay and energy consumption; s4, determining an optimization problem; s5, establishing an unloading decision diagram model, and constructing different task diagrams according to different task types; solving an optimal unloading strategy, and according to different task types, applying a maximum matching minimum cost algorithm to obtain the optimal unloading strategy; and S6, executing task unloading. According to the method, the task allocation strategy problem of unloading calculation is solved, the average time delay and energy consumption of the system are reduced, the task processing success rate is improved, and the resource utilization rate of idle equipment is remarkably improved.

Description

Technical Field

[0001] This invention belongs to the technical field of mobile edge computing task optimization, and specifically relates to a task classification and offloading method in a mobile edge computing network assisted by D2D communication. This method addresses the limited computing resources and energy of mobile user devices by constructing a multi-layered computing offloading architecture through D2D communication, mobile edge computing servers, and remote cloud servers. Based on the differentiated requirements for latency and energy consumption, a joint optimization scheme of task classification and graph matching is proposed, aiming to reduce overall system latency, minimize device energy consumption, and improve the utilization of idle resources. Background Art

[0002] With the rapid development of the Internet of Things and artificial intelligence (AI) applications, the amount of data that mobile devices need to process is exploding. While traditional mobile cloud computing (MCC) offers powerful computing and storage capabilities, the long physical distance between the cloud and users results in significant transmission latency, making it difficult to meet the needs of latency-sensitive applications. Mobile edge computing (MEC) effectively reduces transmission latency by deploying computing resources at the edge of the network, but its limited computing resources make it difficult to support the processing demands of large-scale computing tasks.

[0003] Against this backdrop, the introduction of D2D communication technology offers a new approach to addressing these issues. Through D2D communication, mobile users can directly connect to nearby idle devices and offload tasks, which not only expands computing resources but also reduces transmission latency. However, practical applications still face the following challenges: First, the computing power and battery capacity of different user devices vary, resulting in varying latency and energy consumption for completing the same task; second, offloading a large number of tasks to a single node can cause network congestion and affect service quality; finally, balancing latency and energy consumption with limited computing resources remains a pressing issue.

[0004] Traditional task offloading solutions typically adopt a single optimization objective: either minimizing latency or minimizing energy consumption. This makes it difficult to meet the diverse needs of different task types. Furthermore, these solutions often overlook the computing resources of idle mobile devices, resulting in low system resource utilization. Therefore, rationally classifying tasks based on their characteristics and developing efficient task offloading strategies based on MEC and D2D communication technologies are crucial for improving system performance and user experience. Summary of the Invention

[0005] To address these issues, this paper proposes a graph-theory-based edge computing task offloading method. This method first categorizes tasks based on latency and energy consumption, dividing them into three categories: latency-sensitive, energy-sensitive, and mobile. Based on a maximum matching minimum cost graph-theoretic algorithm, it then develops corresponding offloading strategies for each type of task. This reduces the average system latency and energy consumption, improves the task processing success rate, and significantly enhances the resource utilization of idle devices.

[0006] In order to achieve the above objectives, the present invention adopts the following technical solutions:

[0007] A graph-theory-based edge computing task offloading method includes the following steps:

[0008] Step 1: Get the initialization parameters:

[0009] The number of user devices M, for any user device m, obtain its initial computing resources Channel gain g mx , system bandwidth W, uplink transmission power P m , noise power The number of CPU cycles required to process each task c m , effective switch capacitance constant κ;

[0010] Based on initial computing resources Get the computing resources of user device m Based on W, P m 、g mx 、 Calculating the transfer rate of offloaded tasks

[0011] Based on the channel gain g mx , system bandwidth W, uplink transmission power P m , noise power Get the transfer rate of the offload task

[0012] The user equipment task is represented as a six-tuple set Tas m :

[0013]

[0014] Among them, Ld m Indicates the size of the computational data in the task queue of user device m, d m Indicates the size of the task data currently processed by user device m, represents the maximum tolerable delay of user equipment m, represents the maximum energy consumption of user equipment m, represents the threshold energy consumption of user equipment m, cm is the number of CPU cycles required by user device m to process each task; if Ld m =0, the user equipment can be regarded as being in an idle state; otherwise, the user equipment is in a busy state.

[0015] Step 2. Calculate task latency and energy consumption:

[0016] Step 21 is based on d m 、c m 、 The latency of the computation task when it is performed locally Based on κ, d m 、c m Energy consumption of computing tasks when they are performed locally

[0017] Step 22 is based on d m 、 The latency of computing the task from user device m to user device n and energy consumption Based on and Calculate the total delay of D2D offloading and energy consumption

[0018] Step 23 is based on the cellular transmission power and Calculate the energy consumption of user device m offloading tasks to the MEC server

[0019] Step 3. Classify tasks based on latency and energy consumption:

[0020] Step 31: Maximum tolerable delay with user equipment m For comparison, if Greater than Then the task is a delay-sensitive task and is added to the delay-sensitive task adjacency graph G. t middle;

[0021] The latency-sensitive task adjacency graph is a dynamic, multi-attribute graph model. Its core functions include: task classification, distinguishing latency-sensitive tasks and prioritizing them; resource mapping, finding the optimal offloading path and resources for tasks; and collaborative management, handling dependencies and collaborations between multiple nodes.

[0022] Step 32: The energy consumption of the task when it is calculated locally The maximum energy consumption supported by user device m For comparison, if Greater than Then the task is an energy-sensitive task and is added to the energy-sensitive task adjacency graph G. e middle;

[0023] Step 33: Tasks other than time-sensitive tasks and energy-sensitive tasks are defined as maneuverable tasks and added to the maneuverable task graph G. v middle;

[0024] Step 4: Establish an offloading decision graph model for delay-sensitive, energy-sensitive, and maneuverable tasks, construct delay-sensitive graphs, energy-sensitive graphs, and maneuverable task graphs, and apply the maximum matching minimum cost algorithm to obtain the optimal offloading strategy.

[0025] Preferably, in step 1, the computing resources of user device m are calculated using the following formula: and the transfer rate of the offloaded task

[0026]

[0027] Among them, θ represents the proportion of background services and computing resources that cannot be uninstalled by the device to the total computing resources. represents the initial computing resources of user device m.

[0028] Preferably, in step 21, the time delay of the task when it is calculated locally is calculated according to the following formula: and energy consumption

[0029]

[0030] In step 22, the delay of the task from user device m to user device n is calculated according to the following formula: and energy consumption

[0031]

[0032]

[0033] in, and denote the transmit power of user equipment m and n respectively; user equipment n is a candidate D2D offloading target of user equipment m;

[0034] The total delay of D2D offloading is calculated as follows: and energy consumption

[0035]

[0036] in represents the computing resources of user device n, cn is the number of CPU cycles required by user device n to process each task.

[0037] In step 23, the energy consumption of user device m when offloading tasks to the MEC server is calculated according to the following formula:

[0038]

[0039] in, is the cellular transmission power.

[0040] Preferably, step 4 specifically includes the following steps:

[0041] Step 41: Calculate the delay for delay-sensitive tasks Delay For the weight value, a delay sensitivity graph is constructed and the maximum matching minimum cost method is used to solve the offloading decision with the minimum total delay;

[0042] Step 42: If the idle user has been selected in step 41, the idle node is deleted from the candidate nodes; if the MEC server resources are selected in step S41, the used resources are subtracted from the MEC server and the MEC server resources are updated;

[0043] Step 43: Calculate energy consumption for energy-sensitive tasks Energy consumption As the weight value, an energy consumption sensitivity graph is constructed, and the maximum matching minimum cost method is used to solve the unloading decision with the minimum total energy consumption;

[0044] Step 44: If the idle user has been selected in step 42, the idle node is deleted from the candidate nodes; if the MEC server resource is selected in step 42, the used resources are subtracted from the MEC server and the MEC server resources are updated;

[0045] Step 45: For mobile tasks, calculate the time delay when offloading the task to user device n. and energy consumption Get delay and energy consumption The trade-off value Using V(m,n) as the weight value, construct the maneuver task graph;

[0046] Step 46: Calculate the local processing delay and energy consumption The trade-off value And compare it with V(m,n) of the mobile task. If Vl(m) is smaller, the task is processed locally, otherwise it is not processed locally.

[0047] Among them, ω1, ω2, ω3, and ω4 are all preset trade-off factors, ω1+ω2=1, ω3+ω4=1;

[0048] Step 47: Use the maximum matching and minimum cost method to solve the offloading decision of the maneuver task;

[0049] Maximum matching means allocating as many tasks as possible to available computing resources (such as edge servers, cloud servers, etc.); minimum cost means minimizing the total offloading cost (such as latency, energy consumption, bandwidth overhead, etc.) while meeting the matching quantity.

[0050] Step 48: Use the MEC server to send the uninstall decision to the user device m, and the user device m performs relevant uninstall processing according to the uninstall decision.

[0051] Preferably, in step 31, the adjacency graph G t Taking delay-sensitive tasks as nodes, the maximum tolerable delay For the edge.

[0052] Preferably, in step 32, the energy-sensitive task adjacency graph G e Taking energy-sensitive tasks as nodes, the maximum energy consumption supported by the user device For the edge.

[0053] Preferably, in step 33, the maneuvering task graph G v The mobile task is taken as the node, and the delay-energy consumption trade-off value of the maximum task offloading supported by the user equipment is taken as the edge.

[0054] Compared with the prior art, the present invention has the following technical effects:

[0055] (1) This paper proposes a graph-theory-based edge computing task offloading method. By classifying tasks into three categories: latency-sensitive, energy-sensitive, and mobile, a targeted offloading strategy is implemented for different types of tasks. By using the maximum matching minimum cost algorithm, latency, energy consumption, and their trade-offs are optimized, effectively reducing network pressure and improving task processing efficiency and user experience quality.

[0056] (2) With the assistance of D2D communication, the present invention can fully utilize idle device resources in the area for collaborative task offloading, reducing dependence on central servers. This method improves the flexibility and robustness of the system, optimizes the allocation of computing resources, effectively reduces the overall energy consumption and latency of the system, enhances the efficiency of collaboration between user devices, and improves the overall performance and resource utilization of the network. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a network model diagram of an edge computing task offloading method based on graph theory;

[0058] Figure 2 This is a flow chart of a task offloading strategy based on maximum matching and minimum cost in a graph theory-based edge computing task offloading method.

[0059] Figure 3 The average delay graph under different user equipment;

[0060] Figure 4 This is the average latency graph under different busy device ratios;

[0061] Figure 5 is the average trade-off value diagram under different user devices;

[0062] Figure 6 The average trade-off value diagram under different busy device ratios;

[0063] Figure 7 This is a graph showing the number of users exceeding the limit under different numbers of user devices;

[0064] Figure 8 Utilization graph of idle user devices under different busy devices;

[0065] Figure 9 The average energy consumption graph under different numbers of user devices;

[0066] Figure 10 The figure shows the average energy consumption under different proportions of busy devices. DETAILED DESCRIPTION

[0067] The following describes the embodiments of the present invention through specific embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0068] Based on the cloud-fog-end architecture, this embodiment Figure 1 As shown, user devices are divided into mobile terminals, fog nodes, and a central cloud. Mobile user devices (such as smart home cameras, smart door locks, and mobile phones) can directly exchange tasks between adjacent devices through D2D communication and can choose to offload tasks to fog nodes or cloud servers. By introducing D2D communication and edge computing, this system model enables tasks to be processed in fog nodes close to users, thereby reducing transmission delays and energy consumption, optimizing network resource allocation, and improving user experience quality. This embodiment makes the following assumptions:

[0069] (1) The tasks of the end devices can be classified into different types according to the latency and energy consumption requirements, and different offloading schemes can be adopted to meet the performance requirements of various tasks to reduce the overall system latency and energy consumption.

[0070] (2) During the task offloading process, the MEC server can prioritize the allocation of computing resources to delay-sensitive tasks to ensure the delay requirements, and other tasks can be offloaded when there are sufficient computing resources.

[0071] (3) When there is no interference between D2D communications and computing resources are sufficient, neighboring devices can receive and process tasks through D2D communication, thereby effectively alleviating the computing resource pressure of the MEC server.

[0072] Specifically, the edge computing task offloading method based on graph theory in this embodiment includes the following steps:

[0073] S1. Initialize system parameters.

[0074] Parameters include: the number of user devices M, for any user device m, obtain its initial computing resources Channel gain g mx , system bandwidth W, uplink transmission power P m , noise power The number of CPU cycles required to process each task c m , effective switch capacitance constant k.

[0075] Calculate the computing resources of user device m and the transfer rate of the offloaded task

[0076]

[0077] Among them, θ represents the proportion of background services and computing resources that cannot be uninstalled by the device to the total computing resources. Represents the initial computing resources of user device m

[0078] The user equipment task is represented as a six-tuple set Tas m :

[0079]

[0080] Among them, Ld m (unit: bit) represents the size of the computational data in the task queue of user device m, d m Indicates the size of the task data currently processed by user device m, represents the maximum tolerable delay of user equipment m, represents the maximum energy consumption of user equipment m, represents the threshold energy consumption of user equipment m, c m Is the number of CPU cycles required to process each task. m =0, the user equipment can be regarded as being in an idle state; otherwise, the user equipment is in a busy state.

[0081] The remaining computing resources of the MEC server f mec :

[0082]

[0083] in Indicates the initial computing resources of the MEC server, u m It is a binary data used to indicate whether the user device m occupies the MEC server computing resources. Indicates the MEC server computing resources occupied by user device m.

[0084]

[0085] in, is the time delay for the task to be transmitted from user device m to user device n.

[0086]

[0087] S2 performance analysis, calculating task latency and energy consumption;

[0088] Furthermore, step S2 includes the steps of:

[0089] S2.1 calculates the local task latency and energy consumption;

[0090] S2.2 calculates the D2D offloading delay and energy consumption;

[0091] S2.3 calculates MEC offloading latency and energy consumption;

[0092] S2.4 Calculate the remote cloud offloading latency and energy consumption.

[0093] And in step S2.1:

[0094] The latency of the task when it is computed locally and energy consumption

[0095]

[0096] In step S2.2:

[0097] UE n can be used as a candidate D2D offloading target for UE m. The delay of task transmission from UE m to UE n is and energy consumption

[0098]

[0099] in, and represent the transmit power of user equipment m and n respectively.

[0100] Total latency of D2D offloading and energy consumption

[0101]

[0102] in Represents the computing resources of user device n.

[0103] In step S2.3:

[0104] The delay of user device m offloading tasks to the MEC server and energy consumption

[0105]

[0106] in, is the cellular transmission power.

[0107] In step S2.4:

[0108] Energy consumption of user device m offloading tasks to the remote cloud

[0109]

[0110] In the present invention, it is assumed that delay-sensitive tasks and mobility tasks will not be offloaded to the cloud center, so the delay of offloading to the remote cloud is not considered.

[0111] S3. Task classification based on latency and energy consumption.

[0112] S31, the delay when the user device m offloads the task to the local computing Maximum tolerable delay with user equipment m For comparison, if Greater than The task is considered as a delay-sensitive task, and user device m is added to G t and delete the tasks from the unclassified tasks, where G t It is a delay-sensitive task adjacency graph, with delay-sensitive tasks as nodes and the maximum tolerable delay as For the edge.

[0113] The adjacency graph of latency-sensitive tasks is often used to describe dependencies and latency constraints between tasks, especially in distributed computing, real-time systems, or network scheduling. The graph structure is defined as follows:

[0114] Node,represents an object: each node represents a task (such as a computing task, a data transmission task, etc.).

[0115] Node attributes may include: the computational effort of the task, the data volume (e.g., the number of bytes transferred), the task deadline or maximum tolerable delay, and the task priority (e.g., the criticality level in a real-time system).

[0116] Edges represent relationships: Directed edges represent dependencies or latency constraints between tasks. Dependency: The edge A→BA→B indicates that task BB requires the output of task AA before it can begin. Latency constraint: The weight on the edge represents the maximum allowable latency between two tasks (for example, the latency from A→BA→B must be ≤50ms). The latency-sensitive task adjacency graph is a dynamic, multi-attribute graph model. Its core functions include: task classification, distinguishing and prioritizing latency-sensitive tasks; resource mapping, finding the optimal offloading path and resources for tasks; and collaborative management, handling dependencies and collaboration between multiple nodes.

[0117] S32, energy consumption of user device m task during local calculation The maximum energy consumption supported by the user device (usually a fixed value) for comparison, if Greater than The task is considered to be energy-sensitive and the user device m is added to G e and delete the tasks from the unclassified tasks, where G e It is an energy-sensitive task adjacency graph; with energy-sensitive tasks as nodes, the maximum energy consumption supported by the user equipment is For the edge;

[0118] S33: The remaining unclassified tasks are considered as mobile tasks. v Among them, G v It is a mobile task graph; the mobile tasks are nodes, and the delay-energy consumption trade-off value of the maximum task offloading supported by the user equipment is an edge.

[0119] S4. Determine the optimization problem. The result of this step is not used to generate the unloading decision in step S5, but is used to generate some drawings.

[0120] For delay-sensitive tasks, it is necessary to minimize the execution time within the delay limit. The minimized execution time is obtained by the following formula: Figure 3 and Figure 4In the figure, the delay on the vertical axis is obtained by minimizing the execution time.

[0121]

[0122] For energy-sensitive tasks, it is necessary to minimize the energy consumption of the device. Minimizing the energy consumption of the device is obtained by the following formula: Figure 9 and Figure 10 In the figure, the energy consumption on the vertical axis is obtained by minimizing the energy consumption of the equipment.

[0123]

[0124] For mobile missions, the trade-off between delay and energy consumption needs to be minimized. The trade-off between delay and energy consumption is obtained by the following formula: Figure 5 and Figure 6 In the figure, the delay-energy trade-off value on the vertical axis is obtained based on minimizing the delay-energy trade-off value.

[0125]

[0126] where u m Indicates whether the task is delay-sensitive, 1 indicates sensitive, 0 indicates insensitive; v m Indicates whether the task is energy-sensitive, 1 indicates sensitive, and 0 indicates insensitive; k m Indicates whether the task is offloaded to the MEC server for execution. 1 indicates yes, and 0 indicates no. ω1 and ω2 are trade-off factors that balance the latency and energy consumption of the mobility task. ω1+ω2=1. When the total delay of D2D offloading is or total energy consumption Taken into account.

[0127] This results in the overall optimization objective function within the region, as shown below:

[0128]

[0129] S5. Establish an offloading decision graph model and construct different task graphs according to different task types; solve the optimal offloading strategy and apply the maximum matching minimum cost algorithm according to different task types to obtain the optimal offloading strategy.

[0130] S51. For delayed tasks, calculate the delay Using delay as the weight value, a complete delay-sensitivity graph is constructed and the maximum matching minimum cost method is used to solve the offloading decision with the minimum total delay.

[0131] S52. If the idle user has been selected in S51, the idle node is deleted from the candidate nodes; if the MEC server resources are selected in S51, the used resources are subtracted from the MEC server and the MEC server resources are updated;

[0132] S53. For energy-consuming tasks, calculate energy consumption Energy consumption As the weight value, a complete energy consumption sensitivity graph is constructed and the maximum matching minimum cost method is used to solve the unloading decision with the minimum total energy consumption;

[0133] S54. If the idle user has been selected in S52, the idle node is deleted from the candidate nodes; if the MEC server resources are selected in S52, the used resources are subtracted from the MEC server and the MEC server resources are updated;

[0134] S55. Calculate the delay for mobile missions and energy consumption Calculating latency and energy consumption The trade-off value Using the delay-energy trade-off value V(m,n) as the weight value, a complete maneuvering task graph is constructed;

[0135] S56. Calculate the delay of local processing and energy consumption The trade-off value Vl(m), This value is then compared with the delay-energy tradeoff value V(m,n) for offloading the mobile task. If the delay-energy tradeoff value for local processing is smaller, all weights connected to the user device are set to infinity, indicating that the task is processed locally. If the delay-energy tradeoff value for local processing is larger, the weights remain unchanged. ω1, ω2, ω3, and ω4 are all preset tradeoff factors: ω1+ω2=1, ω3+ω4=1.

[0136] S57, using the maximum matching minimum cost method to solve the unloading decision of the maneuver task;

[0137] Maximum matching means allocating as many tasks as possible to available computing resources (such as edge servers, cloud servers, etc.); minimum cost means minimizing the total offloading cost (such as latency, energy consumption, bandwidth overhead, etc.) while meeting the matching quantity.

[0138] S58. The MEC server sends the uninstallation decision to the user device m, and the user device m performs relevant uninstallation processing according to the uninstallation decision.

[0139] S6. Execute task uninstallation.

[0140] In the field of task offloading, latency sensitivity graphs, energy sensitivity graphs, and maneuvering task graphs are existing tools for optimizing task allocation strategies. The following is a description of the relevant references for their construction methods:

[0141] 1. Timing-Sensitive Graph

[0142] The delay sensitivity graph is used to model the delay constraints of each node (such as mobile devices and edge servers) during task offloading. The goal is to find a task allocation path that meets the deadline. Its core parameters include:

[0143] Task attributes: computational complexity, data volume, and deadline.

[0144] Network status: transmission bandwidth, link latency, and node processing capacity.

[0145] Graph modeling: Mobile devices and edge servers are modeled as nodes in a graph, and task offload paths are modeled as edges. The weight of an edge is the end-to-end latency, which includes local computation latency, transmission latency, and remote computation latency.

[0146] Weight calculation: Delay formula: Ttotal = Tlocal + Ttrans + Tremote

[0147] Tlocal: local computing latency (proportional to task complexity).

[0148] Ttrans: Transmission delay (related to data volume and channel quality).

[0149] Tremote: Remote server processing latency (related to server load).

[0150] Path selection: Use the shortest path algorithm (such as the Dijkstra algorithm) to select the offloading path with the shortest delay.

[0151] References:

[0152] Shi, W., et al. (2016). "Edge Computing: Vision and Challenges." IEEE IoTJournal.

[0153] This paper proposes a latency model and edge node collaboration framework in task offloading.

[0154] Mao,Y.,et al.(2017)."A Survey on Mobile Edge Computing:TheCommunication Perspective."IEEE Communications Surveys&Tutorials.

[0155] The communication model of delay-sensitive task offloading is analyzed in detail.

[0156] 2. Energy-Sensitive Graph

[0157] The energy sensitivity graph is used to optimize the energy consumption during task offloading, with the goal of finding the task allocation solution with the lowest total energy consumption. Its core parameters include:

[0158] Device energy consumption model: local computing energy consumption and transmission energy consumption.

[0159] Network status: channel quality, transmission power.

[0160] Graph modeling: Similar to the delay-sensitive graph, but the edge weight is energy consumption instead of delay.

[0161] Weight calculation: Local computing energy consumption: Elocal = C k / f2

[0162] C: calculation amount, k: constant, f: CPU frequency.

[0163] Transmission energy consumption: Etrans=Ptx·B·dα

[0164] Ptx: transmit power, B: data volume, d: distance, α: path loss exponent.

[0165] Path selection: Use dynamic programming or heuristic algorithms (such as genetic algorithms) to find the path with the lowest energy consumption.

[0166] References:

[0167] Chen, Z., et al. (2018). "Energy-Efficient Task Offloading for MobileEdge Computing with Time-Varying Channels." IEEE Transactions on WirelessCommunications. A task offloading optimization framework based on energy consumption model is proposed.

[0168] Li,X.,et al.(2019)."Energy-Aware Task Offloading in Mobile Edge Computing:A Game-Theoretic Approach."IEEE Communications Letters.Optimizing energy-sensitive task allocation using game theory model.

[0169] 3. Dynamic Task Graph

[0170] The mobile task graph is used for task offloading in dynamic environments (such as device mobility and network topology changes). It requires real-time adjustment of task allocation strategies to balance latency and energy consumption. Its core parameters include:

[0171] Dynamic weight: real-time trade-off coefficient between latency and energy consumption.

[0172] Context awareness: device location, network status changes.

[0173] Multi-objective optimization: Define the comprehensive cost function: W = αT + βE

[0174] α, β: Weight coefficients of latency and energy consumption (dynamically adjusted according to user needs).

[0175] Graph update mechanism: Monitor network status (such as channel quality and node load) in real time and dynamically update edge weights. Use reinforcement learning (RL) or model predictive control (MPC) to generate adaptive policies.

[0176] References:

[0177] Liu, Y., et al. (2020). "Dynamic Task Offloading for Mobile Edge Computing: A Reinforcement Learning Approach." IEEE Transactions on Parallel and Distributed Systems. Propose a dynamic task offloading framework based on reinforcement learning.

[0178] Wang,H.,et al.(2021)."Context-Aware Task Offloading in Mobile Edge Computing:A Survey."IEEE Internet of Things Journal. This paper reviews context-aware methods for task offloading in dynamic environments.

[0179] In order to verify the effect of the present invention, the algorithm was simulated using MATLAB software:

[0180] First, a regular hexagonal cell model and user distribution model are established, and the classification of user devices is simulated based on latency and energy consumption. Then, different offloading strategies are formulated for user devices according to their priority. Figures 3 to 10 By varying the number of users and the proportion of busy devices within a cell, we plotted the energy savings ratio, latency, energy consumption, number of successfully processed users, and idle device utilization using graphs. The experimental results were compared with those of both fully local processing and fully offloaded tasks to the MEC. Finally, by varying the proportion of busy user devices, we obtained a utilization curve for idle devices, revealing the utilization of user devices as communication resources.

[0181] The results show that the proposed algorithm can effectively utilize the computing resources of idle user devices, and the more busy user devices there are, the more obvious the effect is.

Claims

1. A graph-theory-based edge computing task offloading method, characterized in that: The following steps are involved: Step 1: Get the initialization parameters: The number of user devices M, for any user device m, obtain its initial computing resources Channel gain g mx , system bandwidth W, uplink transmission power P m , noise power The number of CPU cycles required to process each task c m , effective switch capacitance constant κ; Based on initial computing resources Get the computing resources of user device m Based on W, P m 、g mx 、 Calculating the transfer rate of offloaded tasks Based on the channel gain g mx , system bandwidth W, uplink transmission power P m , noise power Get the transfer rate of the offload task The user equipment task is represented as a six-tuple set Tas m : Among them, Ld m Indicates the size of the computational data in the task queue of user device m, d m Indicates the size of the task data currently processed by user device m, represents the maximum tolerable delay of user equipment m, represents the maximum energy consumption of user equipment m, represents the threshold energy consumption of user equipment m, c m is the number of CPU cycles required by user device m to process each task; Step 2. Calculate task latency and energy consumption: Step 21 is based on d m 、c m 、 The latency of the computation task when it is performed locally Based on κ, d m 、c m Energy consumption of computing tasks when they are performed locally Step 22 is based on d m 、 The latency of computing the task from user device m to user device n and energy consumption Based on and Calculate the total delay of D2D offloading and energy consumption Step 23 is based on the cellular transmission power and Calculate the energy consumption of user device m offloading tasks to the MEC server Step 3. Classify tasks based on latency and energy consumption: Step 31: Maximum tolerable delay with user equipment m For comparison, if Greater than Then the task is a delay-sensitive task and is added to the delay-sensitive task adjacency graph G. t middle; Step 32: The energy consumption of the task when it is calculated locally The maximum energy consumption supported by user device m For comparison, if Greater than Then the task is an energy-sensitive task and is added to the energy-sensitive task adjacency graph G. e middle; Step 33: Tasks other than time-sensitive tasks and energy-sensitive tasks are defined as maneuverable tasks and added to the maneuverable task graph G. v middle; Step 4: Establish an offloading decision graph model for delay-sensitive, energy-sensitive, and maneuverable tasks, construct delay-sensitive graphs, energy-sensitive graphs, and maneuverable task graphs, and apply the maximum matching minimum cost algorithm to obtain the optimal offloading strategy.

2. The edge computing task offloading method based on graph theory according to claim 1, characterized in that: In step 1, the computing resources of user device m are calculated using the following formula: and the transfer rate of the offloaded task Among them, θ represents the proportion of background services and computing resources that cannot be uninstalled by the device to the total computing resources. represents the initial computing resources of user device m.

3. The edge computing task offloading method based on graph theory according to claim 1, characterized in that: In step 21, the latency of the task when it is computed locally is calculated using the following formula: and energy consumption In step 22, the delay of the task from user device m to user device n is calculated according to the following formula: and energy consumption in, and denote the transmit power of user equipment m and n respectively; user equipment n is a candidate D2D offloading target of user equipment m; The total delay of D2D offloading is calculated as follows: and energy consumption in represents the computing resources of user device n, c n is the number of CPU cycles required by user device n to process each task; In step 23, the energy consumption of user device m when offloading tasks to the MEC server is calculated according to the following formula: in, is the cellular transmission power.

4. The edge computing task offloading method based on graph theory according to claim 1, characterized in that: Step 4 specifically includes the following steps: Step 41: Calculate the delay for delay-sensitive tasks Delay For the weight value, a delay sensitivity graph is constructed and the maximum matching minimum cost method is used to solve the offloading decision with the minimum total delay; Step 42: If the idle user has been selected in step 41, the idle node is deleted from the candidate nodes; if the MEC server resources are selected in step S41, the used resources are subtracted from the MEC server and the MEC server resources are updated; Step 43: Calculate energy consumption for energy-sensitive tasks Energy consumption As the weight value, an energy consumption sensitivity graph is constructed, and the maximum matching minimum cost method is used to solve the unloading decision with the minimum total energy consumption; Step 44: If the idle user has been selected in step 42, the idle node is deleted from the candidate nodes; if the MEC server resource is selected in step 42, the used resources are subtracted from the MEC server and the MEC server resources are updated; Step 45: For mobile tasks, calculate the time delay when offloading the task to user device n. and energy consumption Get delay and energy consumption The trade-off value Using V(m,n) as the weight value, construct the maneuver task graph; Step 46: Calculate the local processing delay and energy consumption The trade-off value And compare it with V(m,n) of the mobile task. If Vl(m) is smaller, the task is processed locally, otherwise it is not processed locally. Among them, ω1, ω2, ω3, and ω4 are all preset trade-off factors, ω1+ω2=1, ω3+ω4=1; Step 47: Use the maximum matching and minimum cost method to solve the offloading decision of the maneuver task; Step 48: Use the MEC server to send the uninstall decision to the user device m, and the user device m performs relevant uninstall processing according to the uninstall decision.

5. The edge computing task offloading method based on graph theory according to claim 1, characterized in that: In step 31, the adjacency graph G t Taking delay-sensitive tasks as nodes, the maximum tolerable delay For the edge.

6. The edge computing task offloading method based on graph theory according to claim 1, characterized in that: In step 32, the energy-sensitive task adjacency graph G e Taking energy-sensitive tasks as nodes, the maximum energy consumption supported by the user device For the edge.

7. The edge computing task offloading method based on graph theory according to claim 1, characterized in that: In step 33, the maneuvering task graph G v The mobile task is taken as the node, and the delay-energy consumption trade-off value of the maximum task offloading supported by the user equipment is taken as the edge.

Images