A method and system for mobile edge computing task offloading and resource allocation

Abstract

This invention discloses a method and system for mobile edge computing task offloading and resource allocation, belonging to the fields of mobile communication and mobile edge computing. The method includes the following steps: acquiring the task characteristics and environmental state of the current time slot, constructing a state vector and inputting it into an offloading decision model to generate candidate offloading decisions; evaluating each candidate decision in terms of uplink transmission, local computing, and edge CPU resource allocation, determining the final offloading decision based on the comprehensive reward value, and updating the model based on the actual execution results. This invention can take into account queue stability, server congestion suppression, and edge energy consumption control. This invention is applicable to multi-edge computing task offloading scenarios.

Description

Technical Field

[0001] This invention relates to the fields of mobile communication and mobile edge computing technology, specifically to a method and system for mobile edge computing task offloading and resource allocation. Background Technology

[0002] With the continuous evolution of 6G networks, smart terminals, and mobile edge computing technologies, a large number of terminal devices are generating tasks that are both computationally intensive and latency-sensitive. Typical applications include augmented reality, video analytics, crowd sensing, autonomous driving collaborative computing, and real-time inference. Traditional cloud-based centralized processing models, due to long backhaul links, severe network congestion, and high response latency, are no longer able to effectively meet the real-time requirements of such tasks. Therefore, mobile edge computing, by offloading computing resources to the base station or access side to provide near-end processing capabilities for terminal tasks, has become an important technical approach to reduce service latency and alleviate pressure on the core network.

[0003] Current task uninstallation methods mostly focus on coarse-grained decision-making levels such as "whether to uninstall" or "which server to uninstall to," and generally suffer from the following shortcomings:

[0004] (1) The decision-making basis mainly relies on the degree of local queue backlog at the terminal or the overall task volume of the server. It lacks the ability to perceive the fine-grained information of "the amount of tasks that a certain user has not yet completed on a certain edge server", which makes it difficult for the system to accurately identify the local task backlog.

[0005] (2) The computation process on the edge server is simply regarded as being completed automatically after uploading, and the allocation of local edge resources is not explicitly included in the joint optimization framework, resulting in a relatively extensive use of edge computing resources;

[0006] (3) In multi-edge server scenarios, there is a lack of proactive suppression mechanisms for server hotspot congestion. Tasks are easily offloaded to a few servers with better channel conditions, which can lead to local congestion.

[0007] (4) Existing solutions usually focus on task completion delay and terminal transmission energy consumption, neglecting the feedback control of edge server computing energy consumption. This can easily lead to long-term high-intensity calls to the edge CPU in pursuit of rapid task clearing, resulting in excessive energy consumption at the edge.

[0008] In summary, there is an urgent need to build a comprehensive technical system that can simultaneously perceive user local task backlog, user-server fine-grained remote backlog, server congestion level and edge energy consumption, and on this basis jointly generate offloading decisions and edge computing power allocation schemes. This has become a key problem that urgently needs to be solved in this field. Summary of the Invention

[0009] To address the challenges of existing task offloading schemes in multi-edge mobile edge computing environments, which struggle to balance local queue stability, server congestion control, and edge computing energy consumption, this invention proposes a mobile edge computing task offloading and resource allocation method and system based on fine-grained remote queue awareness, congestion suppression, joint allocation of edge computing power, and edge energy consumption feedback.

[0010] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for mobile edge computing task offloading and resource allocation, comprising the following steps:

[0011] Step S1: Obtain the task characteristic information of each user terminal in the current time slot;

[0012] Step S2: Obtain the environmental status information of the current time-slot multi-edge computing environment;

[0013] Step S3: Construct the state vector of the current time slot based on the task feature information and the environmental state information;

[0014] Step S4: Input the state vector into the preset unloading decision model to generate at least one candidate unloading decision for the current time slot;

[0015] Step S5: For each candidate offloading decision, evaluate its corresponding uplink transmission process to obtain the uplink transmission evaluation result.

[0016] Step S6: For each candidate unloading decision, allocate local computing resources and evaluate local execution of the local execution task;

[0017] Step S7: For each candidate uninstallation decision, perform edge server CPU resource allocation and edge execution evaluation for the uninstallation execution task;

[0018] Step S8: Based on the uplink transmission evaluation results, local execution evaluation results, and edge execution evaluation results, calculate the comprehensive reward value of each candidate offloading decision, and determine the final offloading decision and corresponding resource allocation result for the current time slot based on the reward value;

[0019] Step S9: Update the system status and the unloading decision model according to the actual execution result of the current time slot, and enter the next time slot to repeat steps S1-S8.

[0020] Furthermore, in step S1, by obtaining the attribute information of the unloaded tasks of each user terminal in the current scheduling time slot, the attribute information is integrated into the task feature vector of the current time slot according to the preset dimensions.

[0021] The attribute information should include at least the amount of input data, the amount of computation required, and the maximum tolerable latency;

[0022] The task feature vector is composed of the amount of input data, the amount of computation required, and the maximum tolerable latency.

[0023] Furthermore, step S2 specifically involves integrating the multi-dimensional environmental state information related to the current unloading task collected by the local scheduler in each scheduling time slot into an environmental state feature vector for the current time slot according to a preset dimension, based on the multi-dimensional environmental state information collected by the local scheduler in each scheduling time slot.

[0024] Environmental status information should include at least:

[0025] The wireless channel status between each user terminal and each mobile edge computing server;

[0026] The backlog of local task queues on each user terminal;

[0027] Current terminal power consumption of each user terminal;

[0028] The fine-grained remote task queue backlog of each user terminal on each edge server is specifically characterized by the amount of unloading tasks that have been offloaded to the edge server but have not yet been processed by the server.

[0029] The current congestion level of each edge server;

[0030] Real-time edge power consumption of each edge server.

[0031] Further, step S5 specifically involves evaluating the uplink transmission from each user terminal to its target edge server for each candidate offload decision; wherein, for each user terminal and its target edge server pairing relationship, an offload utility coefficient is constructed; and based on the offload utility coefficient, the uplink transmission evaluation result corresponding to each candidate offload decision is obtained.

[0032] The following factors should be considered simultaneously when constructing the unloading utility coefficient:

[0033] The level of backlog in the local task queue of this user terminal;

[0034] The degree of backlog in the user terminal's remote task queue on the edge server;

[0035] The current congestion level of the target edge server;

[0036] The power consumption required for the user terminal to perform uplink transmission;

[0037] When the backlog of the local task queue exceeds a preset threshold, the unloading utility coefficient is increased to enhance the unloading tendency; when the backlog of the remote task queue or the server congestion exceeds a preset threshold, the unloading utility coefficient is decreased.

[0038] Furthermore, step S5 specifically includes the following steps:

[0039] Step S51: For each user terminal and its target edge server in the current candidate unloading decision, construct the unloading utility coefficient;

[0040] Step S52: Based on the offloading utility coefficient, calculate the uplink transmission parameters transmitted by each user terminal to the target edge server. The uplink transmission parameters include at least the uplink transmission rate, transmission time, uplink transmission energy consumption, and upload task volume, and obtain the uplink transmission evaluation result corresponding to the current candidate offloading decision.

[0041] Step S53: Traverse each candidate offloading decision and repeat steps S51-S52 above to obtain the uplink transmission evaluation results corresponding to each candidate offloading decision.

[0042] Furthermore, step S6 specifically includes the following steps:

[0043] Step S61: For the local execution task in the current candidate offloading decision, read the local task queue backlog level and the current power consumption status of the user terminal through the local scheduler.

[0044] Step S62: The local scheduler determines the local computing resource allocation parameters based on the read local task queue backlog level and the current power consumption status of the user terminal.

[0045] Based on the local computing resource allocation parameters, calculate the amount of local processing tasks that the user terminal can complete in the current time slot, and estimate the corresponding local computing energy consumption.

[0046] Furthermore, in step S7, for each target edge server, the CPU resource allocation result for that edge server in the current time slot is determined based on the following factors:

[0047] The total number of tasks currently awaiting uninstallation on this edge server;

[0048] The degree of backlog in the remote task queues of each user terminal on the edge server;

[0049] Priority of each pending task;

[0050] The edge power consumption overhead of the edge server;

[0051] The priority of the task to be processed is determined based on at least one of the following: maximum tolerable latency, task waiting time, and local task queue backlog.

[0052] The energy consumption status of the edge server includes the cumulative computing energy consumption level of the edge server in the most recent historical time slots. When the cumulative computing energy consumption level exceeds a preset energy consumption threshold, the allocation of CPU resources to the edge server is reduced.

[0053] Furthermore, step S8 includes the following steps:

[0054] Step S81: For each candidate offloading decision, obtain the uplink transmission evaluation result, local execution evaluation result, and edge execution evaluation result under the candidate offloading decision, and construct the comprehensive evaluation index set corresponding to the candidate offloading decision;

[0055] Step S82: Map the comprehensive evaluation index set to a scalar return value through a preset comprehensive return function, which is used to quantify the comprehensive merits and demerits of the candidate unloading decision in the current time slot;

[0056] Step S83: Select the candidate unloading decision with the largest comprehensive reward value from all candidate unloading decisions as the final unloading decision for the current time slot;

[0057] Step S84: Send the local computing resource allocation parameters, uplink transmission evaluation results and edge server CPU resource allocation results in the current time slot corresponding to the final unloading decision to the corresponding user terminal and target edge server, respectively, so as to execute the task unloading and resource allocation operation in the current time slot.

[0058] Furthermore, step S9 includes the following steps:

[0059] Step S91: After the unloading tasks of each user terminal and edge server in the current time slot are completed, the local scheduler collects the actual task execution results of the current time slot.

[0060] Step S92: Based on the actual task execution results of the current time slot, update the local task queue backlog of each user terminal and the fine-grained remote task queue backlog of each user terminal on each edge server.

[0061] Step S93: Calculate the congestion level of each edge server based on the updated fine-grained remote task queue backlog level;

[0062] Step S94: Update the edge server's energy consumption status based on the actual CPU energy consumption of the current time slot edge server;

[0063] Step S95: The local scheduler generates a state transition sample based on the state vector at the start of the current time slot, the final offloading decision executed in the current time slot, the comprehensive reward value corresponding to the current time slot, and the next state vector updated after the current time slot ends.

[0064] Step S96: When the number of historical samples in the historical experience cache pool reaches the preset sample size threshold, a batch of historical samples are randomly sampled from the historical experience cache pool to perform gradient updates on the network parameters of the unloading decision model. This enables the unloading decision model to adaptively generate task unloading and resource allocation decisions in subsequent time slots based on changes in the user terminal local task queue, fine-grained remote task queue, edge server congestion level, and edge server energy consumption status.

[0065] A mobile edge computing task offloading and resource allocation system that implements the above-described mobile edge computing task offloading and resource allocation method includes:

[0066] The task acquisition module is used to acquire the task feature vector of the current time slot in a multi-edge mobile computing environment;

[0067] The state awareness module communicates with the task acquisition module and is used to acquire the environmental state feature vector of the current time slot;

[0068] The decision generation module, which is connected to the state awareness module, is used to input the state vector of the current time slot, which is constructed based on the task feature vector and the environmental state feature vector of the current time slot, into the preset unloading decision model to generate at least one candidate unloading decision for the current time slot.

[0069] The execution path distribution module divides the tasks to be processed into local execution tasks and unloading execution tasks based on the candidate unloading decision. It sends the local execution tasks to the local computing resource allocation module and the unloading execution tasks to the uplink evaluation module and the edge CPU joint allocation module.

[0070] The uplink evaluation module communicates with the execution path distribution module. It is used to construct the offloading utility coefficient based on the pairing relationship between the user terminal corresponding to the offloading execution task and the target edge server, and to obtain the uplink transmission evaluation result corresponding to each candidate offloading decision.

[0071] The local computing resource allocation module communicates with the execution path distribution module and is used to obtain the amount of local processing tasks that the user terminal can complete in the current time slot and the corresponding local computing energy consumption based on the local execution tasks in the current candidate unloading decision.

[0072] The edge CPU joint allocation module communicates with the execution path distribution module and is used to explicitly allocate the CPU computing resources of each edge server after the unloading task is uploaded to the target edge server for the current candidate unloading decision. It determines the CPU resource allocation result of the edge server in the current time slot and obtains the processing volume and CPU energy consumption of the unloading task completed by the edge server based on the CPU resource allocation result of the edge server in the current time slot.

[0073] The comprehensive evaluation and execution module is connected to the uplink evaluation module, the local computing resource allocation module, and the edge CPU joint allocation module, respectively. It is used to calculate the reward value of each candidate offloading decision based on the comprehensive evaluation index set corresponding to each candidate offloading decision and through a preset comprehensive reward function. Based on the reward value, it determines the final offloading decision of the current time slot and executes the offloading task.

[0074] The strategy update module communicates with the comprehensive evaluation and execution module to update the unloading decision model based on the actual unloading task execution results collected in the current time slot.

[0075] The advantages of this invention over the prior art are as follows: This invention constructs the state vector of the current time slot by obtaining the task feature vector and environmental state feature vector of the current time slot, and inputs the state vector of the current time slot into a preset unloading decision model to generate at least one candidate unloading decision for the current time slot; and simultaneously completes the uplink unloading task transmission allocation, local computing resource allocation and edge server CPU resource allocation for each candidate unloading decision, thereby ensuring the stable execution of the unloading task while suppressing edge server overload and excessive energy consumption, and realizing the coordinated optimization of task unloading performance and edge end energy consumption. Attached Figure Description

[0076] The present invention will be further described below with reference to the accompanying drawings:

[0077] Figure 1 This is a schematic flowchart of the method of the present invention;

[0078] Figure 2 This is a schematic diagram of the system structure of the present invention;

[0079] Figure 3 This is a schematic diagram illustrating the closed-loop interaction relationship between the user terminal-side state, the edge server-side state, the wireless channel state, and the execution result feedback of the present invention. Detailed Implementation

[0080] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate relative orientations or positional relationships and are used only for the convenience of describing the invention and 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, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0081] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0082] like Figures 1 to 3 As shown, this invention provides a method for mobile edge computing task offloading and resource allocation. By simultaneously constructing the terminal's local task queue, terminal energy consumption, fine-grained remote queues of users on each edge server, server congestion levels, and edge energy consumption overhead in each scheduling time slot, candidate offloading decisions are generated. For each candidate offloading decision, local computing resource allocation, uplink transmission resource allocation, and edge local resource allocation are simultaneously performed. This ensures stable execution of offloading tasks while suppressing edge server overload and excessive energy consumption, achieving coordinated optimization of task offloading performance and edge energy consumption. The method includes the following steps:

[0083] Step S1: In a multi-edge mobile computing environment, obtain the attribute information of the unloaded tasks of each user terminal in the current scheduling time slot, and integrate the attribute information into the task feature vector of the current time slot according to the preset dimensions.

[0084] Specifically, in a mobile edge computing scenario with multiple servers deployed, the local scheduler obtains the attribute information of the tasks to be unloaded by each user terminal at the beginning of each scheduling time slot. The attribute information includes at least the amount of input data, the amount of computation required, and the maximum tolerable latency. The scheduler integrates the attribute information into a task feature vector for the current time slot according to a preset dimension. The task feature vector is composed of the amount of input data, the amount of computation required, and the maximum tolerable latency.

[0085] Step S2: In each scheduling time slot, the local scheduler collects multi-dimensional environmental state information related to the current unloading task and integrates the multi-dimensional environmental state information into an environmental state feature vector for the current time slot according to preset dimensions.

[0086] Environmental status information should include at least:

[0087] (1) The wireless channel status between each user terminal and each mobile edge computing server;

[0088] (2) The backlog of local task queues (backlog length) of each user terminal.

[0089] (3) Current terminal power consumption (power consumption overhead) of each user terminal.

[0090] (4) The degree of backlog of fine-grained remote task queues of each user terminal on each edge server. The degree of backlog of fine-grained remote task queues is specifically characterized by the amount of unloading tasks that user terminals have been unloaded to the edge server but have not yet been processed by the server.

[0091] (5) The current congestion level of each edge server (global congestion level, such as its total task queue length or service rate percentage).

[0092] (6) Real-time edge power consumption of each edge server, such as computing power.

[0093] Step S3: Concatenate the task feature vector and the environment state feature vector of the current time slot according to the feature dimension to construct the state vector of the current time slot.

[0094] Step S4: Input the state vector of the current time slot into the preset unloading decision model to generate at least one candidate unloading decision for the current time slot. The unloading decision model is used to generate the corresponding unloading decision result based on the current time slot state.

[0095] Specifically, the candidate uninstallation decision is used to instruct each user terminal to execute the current uninstallation task locally or to uninstall it to a target edge server for execution. For uninstallation tasks with a decision confidence level lower than a preset threshold, the uninstallation decision module generates multiple candidate uninstallation decisions for subsequent steps to allocate resources and select the best option, thereby improving the reliability of uninstallation in a dynamic environment.

[0096] Step S5: For each candidate offloading decision, evaluate its corresponding uplink transmission process to obtain an uplink transmission evaluation result. Specifically, for each candidate offloading decision, evaluate the uplink transmission of each user terminal to its target edge server based on a partial environmental state feature vector (in this embodiment, the partial environmental state feature vector includes the backlog of the user terminal's local task queue, the backlog of the user terminal's fine-grained remote task queue on the target edge server, the congestion level of the target edge server, and the energy consumption of the user terminal for uplink transmission to the target edge server). Specifically, for each user terminal and its target edge server pairing relationship, construct an offloading utility coefficient; based on the offloading utility coefficient, obtain the uplink transmission evaluation result corresponding to each candidate offloading decision.

[0097] Step S51: Consider at least the following factors simultaneously to construct the unloading utility coefficient:

[0098] (1) The degree of backlog in the local task queue of the user terminal;

[0099] (2) The degree of backlog in the fine-grained remote task queue of the user terminal on the target edge server;

[0100] (3) The current congestion level of the target edge server;

[0101] (4) The energy consumption required for the user terminal to transmit uplink data to the target edge server.

[0102] Specifically, when the backlog of the local task queue exceeds a preset threshold, the unloading utility coefficient is increased to enhance the unloading tendency; when the backlog of the remote task queue or the server congestion exceeds a preset threshold, the unloading utility coefficient is decreased to suppress the continued unloading to that server, thereby avoiding local task accumulation and server hotspot congestion.

[0103] Step S52: Based on the offloading utility coefficient, further calculate the uplink transmission rate, transmission time, uplink transmission energy consumption and upload task volume of each user terminal to the target edge server, and obtain the uplink transmission evaluation result corresponding to the current candidate offloading decision.

[0104] In one alternative implementation, let the set of user terminals be... The edge server set is The current scheduling slot is The time slot length is User terminal The local task queue backlog is denoted as User terminal On the edge server The degree of backlog in the fine-grained remote task queue is denoted as Edge server The level of congestion is denoted as User terminal With edge servers The wireless channel gain between them is denoted as User terminal To edge servers The energy consumption for uplink transmission is denoted as ;

[0105] For user terminals With edge servers The pairing relationship and the unloading utility coefficient can be expressed as:

[0106] ;

[0107] in, This indicates the degree of backlog in the local task queue of user terminal i after normalization.

[0108] Represents the normalized user terminal On the edge server Fine-grained remote task queue backlog levels;

[0109] Represents the normalized edge server The degree of congestion;

[0110] Represents the normalized user terminal To edge servers Energy consumption for uplink transmission;

[0111] These are non-negative weight parameters.

[0112] When the backlog of the local task queue on the user terminal increases, the offloading utility coefficient increases to enhance the offloading tendency; when the backlog of the fine-grained remote task queue on the target edge server, the congestion level of the edge server, or the uplink transmission energy consumption increases, the offloading utility coefficient decreases to suppress the continued offloading of tasks to the target edge server.

[0113] User terminal To edge servers Uplink transmission rate It can be represented as:

[0114] ;

[0115] in, Indicates allocation to user terminals With edge servers Uplink bandwidth, Indicates user terminal The transmission power, Indicates noise power. This indicates the interference power.

[0116] User terminals in the current time slot Upload to edge server workload It can be represented as:

[0117] ;

[0118] Where min{·} represents taking the smaller of the two values. Indicates user terminal The backlog level of the local task queue in the current time slot. This indicates that user terminal i can upload up to the edge server via the uplink within the current time slot. The workload.

[0119] User terminal To edge servers Time required for uplink transmission It can be represented as:

[0120] ;

[0121] Corresponding uplink transmission power consumption It can be represented as:

[0122] ;

[0123] Step S53: Traverse each candidate unloading decision and repeat steps S51-S52 above to obtain the uplink evaluation results corresponding to each candidate unloading decision, which are used for the comprehensive evaluation of subsequent candidate unloading decisions and the selection of the optimal unloading decision.

[0124] To facilitate differentiation between different candidate unloading decisions, the candidate unloading decision subscripts are omitted from the formulas in steps S51 to S52. For the... Each of the candidate offloading decisions is denoted by its corresponding uplink transmission rate, upload task volume, transmission time, and transmission energy consumption as follows: , , and This serves as the input for local computing resource allocation, edge server CPU resource allocation, and overall reward calculation under the candidate offloading decision.

[0125] Step S6: For each local execution task in the candidate offloading decision, perform local computing resource allocation and local execution evaluation. Specifically, the local scheduler determines the local computing resource allocation parameters based on the local execution tasks read from the current candidate offloading decision; based on the local computing resource allocation parameters, it calculates the amount of local processing tasks that the user terminal can complete in the current time slot and the corresponding local computing energy consumption.

[0126] Specifically, the following steps are included:

[0127] Step S61: For the local execution task in the current candidate offloading decision, read the local task queue backlog level and the current power consumption status of the user terminal through the local scheduler.

[0128] Step S62: The local scheduler determines the local computing resource allocation parameters based on the read local task queue backlog level and the current power consumption status of the user terminal; the local computing resource allocation parameters include at least the local CPU operating frequency or an equivalent local computing capacity allocation value.

[0129] Based on local computing resource allocation parameters, the amount of local processing tasks that the user terminal can complete in the current time slot is calculated, and the corresponding local computing energy consumption is estimated. This local computing energy consumption will be used as the local execution cost of this candidate offloading decision, and will be used for the comprehensive evaluation of subsequent candidate offloading decisions and the selection of the optimal offloading decision.

[0130] In one alternative implementation, a user terminal is provided. The number of CPU cycles required to process a unit of task data is , No. Under the candidate uninstallation decision, the user terminal The allocated local CPU operating frequency is User terminal The maximum local CPU operating frequency is Then the local CPU operating frequency satisfies:

[0131]

[0132] User terminal The amount of local processing tasks that can be completed within the current time slot is:

[0133]

[0134] The corresponding local computing power consumption is:

[0135]

[0136] in, Indicates user terminal The energy consumption coefficient is calculated. When the first... The candidate uninstallation decision did not include the user terminal. When the task is determined to be executed locally, let The corresponding local processing workload and local computing power consumption are both zero. This results in... and Together constitute the first The local execution evaluation results of each candidate unloading decision are used to calculate the subsequent comprehensive return value.

[0137] Step S7: For each candidate unloading decision's unloading execution task, perform edge server CPU resource allocation and edge execution evaluation. Specifically, for the unloading task executed by the current candidate unloading decision, after the unloading task is uploaded to the target edge server, the CPU computing resources of each edge server are explicitly allocated to determine the CPU resource allocation result of the edge server in the current time slot. Based on the CPU resource allocation result of the edge server in the current time slot, the processing volume and CPU energy consumption of the unloading task completed by the edge server are obtained.

[0138] For each target edge server, the CPU resource allocation result for that edge server in the current time slot is determined based on the following factors:

[0139] (1) The total number of tasks currently awaiting unloading on the edge server;

[0140] (2) The degree of backlog in the fine-grained remote task queue of each user terminal on the edge server;

[0141] (3) Priority of each task to be uninstalled;

[0142] (4) The edge power consumption of the edge server;

[0143] The priority of the pending tasks is determined based on at least one of the following: maximum tolerable latency, task waiting time, and local task queue backlog.

[0144] The energy consumption status of the edge server includes the cumulative computing energy consumption level of the edge server in the most recent historical time slots. When the cumulative computing energy consumption level exceeds the preset energy consumption threshold, the allocation of CPU resources to the edge server is reduced to suppress its continuous rise in energy consumption and realize closed-loop control of edge energy consumption.

[0145] In one alternative implementation, for the first One candidate offload decision, edge server User terminal The current number of tasks that can be processed is:

[0146]

[0147] in, Indicates the user terminal at the start of the current time slot. On the edge server The amount of unprocessed tasks, Indicates the first Under the candidate unloading decision, the current time slot is newly uploaded to the edge server. The workload.

[0148] Set up an edge server In the The candidate uninstallation decision is assigned to the user terminal. The local CPU operating frequency is Edge server The computing resource capacity is The edge server CPU resource allocation result satisfies:

[0149]

[0150] Edge server For the user terminal in the current time slot The completed uninstallation task processing volume is:

[0151]

[0152] Edge server In the The CPU energy consumption generated under each candidate unloading decision is:

[0153]

[0154] in, Represents edge server The energy consumption coefficient is calculated. It is influenced by the fine-grained remote task queue backlog at each user terminal, the priority of tasks to be processed, and the energy consumption status of the edge server. The value of is determined by the fact that when the cumulative historical energy consumption of the edge server exceeds a preset energy consumption threshold, the local CPU operating frequency is reduced within the CPU resource constraints mentioned above to suppress the continuous increase in the edge server's computing energy consumption. This results in... and Together constitute the first Edge execution evaluation results of candidate unloading decisions.

[0155] Step S8: Based on the uplink transmission evaluation results, local execution evaluation results, and edge execution evaluation results, calculate the comprehensive reward value of each candidate offloading decision, and determine the final offloading decision and corresponding resource allocation result for the current time slot based on the reward value. Based on the uplink transmission evaluation results, local execution evaluation results, and edge execution evaluation results corresponding to each candidate offloading decision, construct a comprehensive evaluation index set corresponding to each candidate offloading decision. Based on the comprehensive evaluation index set corresponding to each candidate offloading decision, calculate the reward value of each candidate offloading decision using a preset comprehensive reward function. Determine the final offloading decision for the current time slot based on the reward value to perform task offloading and resource allocation tasks.

[0156] Specifically, the following steps are included:

[0157] Step S81: For each candidate offloading decision, obtain the uplink transmission evaluation result, local execution evaluation result, and edge execution evaluation result under the candidate offloading decision, and construct the comprehensive evaluation index set corresponding to the candidate offloading decision;

[0158] Step S82: Map the comprehensive evaluation index set to a scalar return value through a preset comprehensive return function, which is used to quantify the comprehensive merits and demerits of the candidate unloading decision in the current time slot;

[0159] In one alternative implementation, for the first The overall return value of each candidate unloading decision It can be represented as:

[0160]

[0161] in, Indicates the first After the user terminal executes the candidate unload decision The degree of backlog in the local task queue, Indicates the first After the user terminal executes the candidate unload decision On the edge server Fine-grained remote task queue backlog level Represents edge server The degree of congestion, Indicates user terminal Local computing power consumption, Represents edge server CPU power consumption Indicates user terminal To edge servers Energy consumption for uplink transmission These are the corresponding non-negative weight parameters.

[0162] In calculating the first When considering the overall return value of each candidate unloading decision, and This represents the prediction queue state corresponding to the candidate unloading decision. Among them, the first... Local processing workload for each candidate unloading decision Upload task volume and edge processing workload Substituting the queue update relationship described in step S92, we obtain and Then, substitute the predicted fine-grained remote task queue backlog into the congestion calculation relationship described in step S93 to obtain... .

[0163] The predicted queue status and predicted congestion level are only used for the comprehensive reward comparison between different candidate offloading decisions in the current time slot, and do not indicate that the system has actually executed the candidate offloading decision. After determining the final offloading decision, the system updates the real local task queue, fine-grained remote task queue, server congestion level, and edge server energy consumption status according to steps S92 to S94 based on the actual execution result of the final offloading decision.

[0164] As shown in the comprehensive reward function above, the smaller the local task queue backlog, the fine-grained remote task queue backlog, the edge server congestion, the terminal local computing power consumption, the edge server CPU power consumption, and the uplink transmission power consumption, the larger the comprehensive reward value corresponding to the candidate offloading decision. Step S83: Select the candidate offloading decision with the largest comprehensive reward value from all candidate offloading decisions as the final offloading decision for the current time slot; the final offloading decision for the current time slot is expressed as:

[0165] ;

[0166] in, This represents the set of candidate offloading decisions generated in the current time slot. Indicates the first One candidate uninstallation decision This indicates the final unloading decision for the current time slot.

[0167] Step S83: Select the candidate unloading decision with the largest comprehensive reward value from all candidate unloading decisions as the final unloading decision for the current time slot;

[0168] Step S84: Send the local computing resource allocation parameters, uplink transmission evaluation results and edge server CPU resource allocation results in the current time slot corresponding to the final unloading decision to the corresponding user terminal and target edge server, respectively, so as to execute the task unloading and resource allocation operation in the current time slot.

[0169] The comprehensive evaluation should be based on at least the following three categories of indicators:

[0170] (1) Computation and transmission resource overhead, including local processing task volume, uplink offloading task transmission results, and edge server offloading task processing volume;

[0171] (2) The state evolution results include the local task queue backlog of each user terminal, the fine-grained remote task queue backlog of each user terminal on each edge server, and the global congestion of each edge server.

[0172] (3) Energy consumption, including local computing energy consumption of each user terminal and CPU energy consumption of each edge server.

[0173] Step S9: Update the system state and offloading decision model based on the actual execution results of the current time slot, and then repeat steps S1-S8 in the next time slot. Specifically, based on the collected actual offloading and resource allocation task execution results of the current time slot, update the local task queue backlog of user terminals, the fine-grained remote task queue backlog of each user terminal on each edge server, the congestion level and energy consumption status of each edge server, and update the offloading decision model based on the state transition samples before repeating steps S1-S8 in the next time slot.

[0174] Step S9 specifically includes the following steps:

[0175] Step S91: At the end of the current time slot, the local scheduler collects the actual task execution results of the current time slot; the actual offloading task execution results include at least the actual local processing task volume of each user terminal, the actual upload task volume of each user terminal to the target edge server, the actual offloading task processing volume completed by each edge server, the actual uplink transmission energy consumption of each user terminal, the actual local computing energy consumption of each user terminal, and the actual CPU energy consumption of each edge server.

[0176] In one alternative implementation, the user terminal The actual terminal power consumption generated in the current time slot consists of local computing power consumption and uplink transmission power consumption, which can be expressed as:

[0177]

[0178] in, This indicates the user terminal after the final unloading decision is actually executed. The resulting local computing power consumption, Indicates user terminal To edge servers The actual transmission energy consumption generated by the upload task Indicates user terminal The actual terminal energy consumption in the current time slot is used to construct the user terminal energy consumption status for the next time slot.

[0179] Step S92: Based on the actual task execution results of the current time slot, update the local task queue backlog of each user terminal and the fine-grained remote task queue backlog of each user terminal on each edge server.

[0180] In one alternative implementation, the user terminal Local task queue backlog Update as follows:

[0181] ;

[0182] in, Indicates user terminal The backlog level of the local task queue in the current time slot. Indicates user terminal The amount of new tasks added in the current time slot Indicates user terminal The amount of local processing tasks completed in the current time slot, Indicates user terminal Uploaded to the edge server in the current time slot The workload.

[0183] User terminal On the edge server Fine-grained remote task queue backlog Update as follows:

[0184] ;

[0185] in, Indicates user terminal On the edge server The current number of tasks that have not yet been processed. Represents edge server User terminals completed within the current time slot The amount of unloading tasks processed.

[0186] Step S93: Calculate the congestion level of each edge server based on the updated fine-grained remote task queue backlog. In one optional implementation, the edge servers... congestion level It can be represented as:

[0187] ;

[0188] in, Represents edge server The computing resource capacity, To prevent constants with a denominator of zero.

[0189] Step S94: Update the edge server's energy consumption status based on the actual CPU energy consumption of the current time slot edge server. In one optional implementation, the edge server... Historical cumulative energy consumption level It can be represented as:

[0190] ;

[0191] in, Represents edge server Based on the historical cumulative energy consumption level prior to the current time slot, Represents edge server CPU power consumption generated in the current time slot It is the energy consumption attenuation factor, and .

[0192] Step S95: The local scheduler generates a state transition sample (s(t), a*(t), R(t), s(t+1)) based on the state vector s(t) at the start of the current time slot, the final offloading decision a*(t) executed in the current time slot, the comprehensive reward value R(t) corresponding to the current time slot, and the next state vector s(t+1) updated after the current time slot ends, and stores the state transition sample in the historical experience cache pool.

[0193] Where s(t) represents the state vector at the start of the current time slot, a*(t) represents the final unloading decision of the current time slot, R(t) represents the comprehensive reward value corresponding to the final unloading decision, and s(t+1) represents the next state vector obtained after the current time slot ends.

[0194] Step S96: When the number of historical samples in the historical experience cache pool reaches the preset sample size threshold, a batch of historical samples are randomly sampled from the historical experience cache pool to perform gradient updates on the network parameters of the unloading decision model. This enables the unloading decision model to adaptively generate task unloading and resource allocation decisions in subsequent time slots based on changes in the user terminal local task queue, fine-grained remote task queue, edge server congestion level, and edge server energy consumption status.

[0195] Through the above steps, the unloading decision model is updated by using feedback from actual unloading tasks after each time slot ends. This enables the unloading decisions in subsequent time slots to reduce task backlog while suppressing local server congestion and excessive energy consumption in edge computing, thereby achieving closed-loop adaptive optimization for multi-edge computing environments.

[0196] Steps S1 to S9 of this invention constitute a complete technical path consisting of task information acquisition, multi-dimensional environment perception, candidate offloading decision generation, uplink transmission evaluation, local computing resource allocation, edge local computing resource allocation, comprehensive evaluation of candidate offloading decisions, and closed-loop update of the offloading decision model driven by execution feedback. This path couples offloading decision and resource allocation into an iteratively executable time-series decision process, realizing joint optimization for multi-edge computing environments.

[0197] This invention provides a mobile edge computing task offloading and resource allocation system that implements the above-described mobile edge computing task offloading and resource allocation method, comprising:

[0198] The task acquisition module is used to acquire the task feature vector of the current time slot in a multi-edge mobile computing environment;

[0199] The state awareness module communicates with the task acquisition module and is used to acquire the environmental state feature vector of the current time slot;

[0200] The decision generation module, which is connected to the state awareness module, is used to input the state vector of the current time slot, which is constructed based on the task feature vector and the environmental state feature vector of the current time slot, into the preset unloading decision model to generate at least one candidate unloading decision for the current time slot.

[0201] The execution path distribution module communicates with the decision generation module and is used to divide the tasks to be processed into local execution tasks and unloading execution tasks according to the candidate unloading decision. The local execution tasks are sent to the local computing resource allocation module, and the unloading execution tasks are sent to the uplink evaluation module and the edge CPU joint allocation module.

[0202] The uplink evaluation module communicates with the execution path distribution module. It is used to construct the offloading utility coefficient based on the pairing relationship between the user terminal corresponding to the offloading execution task and the target edge server, and to obtain the uplink transmission evaluation result corresponding to each candidate offloading decision.

[0203] The local computing resource allocation module communicates with the execution path distribution module and is used to obtain the amount of local processing tasks that the user terminal can complete in the current time slot and the corresponding local computing energy consumption based on the local execution tasks in the current candidate unloading decision.

[0204] The edge CPU joint allocation module communicates with the execution path distribution module and is used to explicitly allocate the CPU computing resources of each edge server after the unloading task is uploaded to the target edge server for the current candidate unloading decision. It determines the CPU resource allocation result of the edge server in the current time slot and obtains the processing volume and CPU energy consumption of the unloading task completed by the edge server based on the CPU resource allocation result of the edge server in the current time slot.

[0205] The comprehensive evaluation and execution module is connected to the uplink evaluation module, the local computing resource allocation module, and the edge CPU joint allocation module, respectively. It is used to calculate the reward value of each candidate offloading decision based on the comprehensive evaluation index set corresponding to each candidate offloading decision and through a preset comprehensive reward function. Based on the reward value, it determines the final offloading decision of the current time slot and executes the offloading task.

[0206] The strategy update module communicates with the comprehensive evaluation and execution module to update the unloading decision model based on the actual unloading task execution results collected in the current time slot.

[0207] Regarding the specific structure of this invention, it should be noted that the connection relationships between the various component modules used in this invention are definite and achievable. Except as specifically described in the embodiments, their specific connection relationships can bring about corresponding technical effects and solve the technical problems proposed by this invention without relying on the execution of corresponding software programs. The models of the components, modules, and specific components appearing in this invention, the connection methods between them, and the conventional usage methods and expected technical effects brought about by the above technical features, unless specifically described, are all publicly disclosed content in patents, journal articles, technical manuals, technical dictionaries, and textbooks that can be obtained by those skilled in the art before the application date, or belong to conventional technology, common knowledge, and other existing technologies in this field. There is no need to elaborate, which makes the technical solution provided in this case clear, complete, and achievable, and can reproduce or obtain corresponding physical products based on this technical means.

[0208] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for mobile edge computing task offloading and resource allocation, characterized in that, Includes the following steps: Step S1: Obtain the task characteristic information of each user terminal in the current time slot; Step S2: Obtain the environmental status information of the current time-slot multi-edge computing environment; Step S3: Construct the state vector of the current time slot based on the task feature information and the environmental state information; Step S4: Input the state vector into the preset unloading decision model to generate at least one candidate unloading decision for the current time slot; Step S5: For each candidate offloading decision, evaluate its corresponding uplink transmission process to obtain the uplink transmission evaluation result. Step S6: For each candidate unloading decision, allocate local computing resources and evaluate local execution of the local execution task; Step S7: For each candidate uninstallation decision, perform edge server CPU resource allocation and edge execution evaluation for the uninstallation execution task; Step S8: Based on the uplink transmission evaluation results, local execution evaluation results, and edge execution evaluation results, calculate the comprehensive reward value of each candidate offloading decision, and determine the final offloading decision and corresponding resource allocation result for the current time slot based on the reward value; Step S9: Update the system status and the unloading decision model according to the actual execution result of the current time slot, and enter the next time slot to repeat steps S1-S8.

2. The method for mobile edge computing task offloading and resource allocation according to claim 1, characterized in that, In step S1, the attribute information of the unloaded tasks of each user terminal in the current scheduling time slot is obtained, and the attribute information is integrated into the task feature vector of the current time slot according to the preset dimensions. The attribute information should include at least the amount of input data, the amount of computation required, and the maximum tolerable latency; The task feature vector is composed of the amount of input data, the amount of computation required, and the maximum tolerable latency.

3. The method for mobile edge computing task offloading and resource allocation according to claim 1, characterized in that, Step S2 specifically involves integrating the multi-dimensional environmental state information related to the current unloading task collected by the local scheduler in each scheduling time slot into an environmental state feature vector for the current time slot according to preset dimensions, based on the multi-dimensional environmental state information collected by the local scheduler in each scheduling time slot. Environmental status information should include at least: The wireless channel status between each user terminal and each mobile edge computing server; The backlog of local task queues on each user terminal; Current terminal power consumption of each user terminal; The fine-grained remote task queue backlog of each user terminal on each edge server is specifically characterized by the amount of unloading tasks that have been offloaded to the edge server but have not yet been processed by the server. The current congestion level of each edge server; Real-time edge power consumption of each edge server.

4. The method for mobile edge computing task offloading and resource allocation according to claim 1, characterized in that, Step S5 specifically involves evaluating the uplink transmission from each user terminal to its target edge server for each candidate offload decision; wherein, for each user terminal and its target edge server pairing relationship, an offload utility coefficient is constructed; and based on the offload utility coefficient, the uplink transmission evaluation result corresponding to each candidate offload decision is obtained. The following factors should be considered simultaneously when constructing the unloading utility coefficient: The level of backlog in the local task queue of this user terminal; The degree of backlog in the user terminal's remote task queue on the edge server; The current congestion level of the target edge server; The power consumption required for the user terminal to perform uplink transmission; When the backlog of the local task queue exceeds a preset threshold, the unloading utility coefficient is increased to enhance the unloading tendency; when the backlog of the remote task queue or the server congestion exceeds a preset threshold, the unloading utility coefficient is decreased.

5. The mobile edge computing task offloading and resource allocation method according to claim 4, characterized in that, Step S5 specifically includes the following steps: Step S51: For each user terminal and its target edge server in the current candidate unloading decision, construct the unloading utility coefficient; Step S52: Based on the offloading utility coefficient, calculate the uplink transmission parameters transmitted by each user terminal to the target edge server. The uplink transmission parameters include at least the uplink transmission rate, transmission time, uplink transmission energy consumption, and upload task volume, and obtain the uplink transmission evaluation result corresponding to the current candidate offloading decision. Step S53: Traverse each candidate offloading decision and repeat steps S51-S52 above to obtain the uplink transmission evaluation results corresponding to each candidate offloading decision.

6. The method for mobile edge computing task offloading and resource allocation according to claim 5, characterized in that, Step S6 specifically includes the following steps: Step S61: For the local execution task in the current candidate offloading decision, read the local task queue backlog level and the current power consumption status of the user terminal through the local scheduler. Step S62: The local scheduler determines the local computing resource allocation parameters based on the read local task queue backlog level and the current power consumption status of the user terminal. Based on the local computing resource allocation parameters, calculate the amount of local processing tasks that the user terminal can complete in the current time slot, and estimate the corresponding local computing energy consumption.

7. The method for mobile edge computing task offloading and resource allocation according to claim 6, characterized in that, In step S7, for each target edge server, the CPU resource allocation result for that edge server in the current time slot is determined based on the following factors: The total number of tasks currently awaiting unloading on this edge server; The degree of backlog in the remote task queues of each user terminal on the edge server; Priority of each pending task; The edge power consumption overhead of the edge server; The priority of the task to be processed is determined based on at least one of the following: maximum tolerable latency, task waiting time, and local task queue backlog. The energy consumption status of the edge server includes the cumulative computing energy consumption level of the edge server in the most recent historical time slots. When the cumulative computing energy consumption level exceeds a preset energy consumption threshold, the allocation of CPU resources to the edge server is reduced.

8. The method for mobile edge computing task offloading and resource allocation according to claim 7, characterized in that, Step S8 includes the following steps: Step S81: For each candidate offloading decision, obtain the uplink transmission evaluation result, local execution evaluation result, and edge execution evaluation result under the candidate offloading decision, and construct the comprehensive evaluation index set corresponding to the candidate offloading decision; Step S82: Map the comprehensive evaluation index set to a scalar return value through a preset comprehensive return function, which is used to quantify the comprehensive merits and demerits of the candidate unloading decision in the current time slot; Step S83: Select the candidate unloading decision with the largest comprehensive reward value from all candidate unloading decisions as the final unloading decision for the current time slot; Step S84: Send the local computing resource allocation parameters, uplink transmission evaluation results and edge server CPU resource allocation results in the current time slot corresponding to the final unloading decision to the corresponding user terminal and target edge server, respectively, so as to execute the task unloading and resource allocation operation in the current time slot.

9. The method for mobile edge computing task offloading and resource allocation according to claim 8, characterized in that, Step S9 includes the following steps: Step S91: After the unloading tasks of each user terminal and edge server in the current time slot are completed, the local scheduler collects the actual task execution results of the current time slot. Step S92: Based on the actual task execution results of the current time slot, update the local task queue backlog of each user terminal and the fine-grained remote task queue backlog of each user terminal on each edge server. Step S93: Calculate the congestion level of each edge server based on the updated fine-grained remote task queue backlog level; Step S94: Update the edge server's energy consumption status based on the actual CPU energy consumption of the current time slot edge server; Step S95: The local scheduler generates a state transition sample based on the state vector at the start of the current time slot, the final offloading decision executed in the current time slot, the comprehensive reward value corresponding to the current time slot, and the next state vector updated after the current time slot ends. Step S96: When the number of historical samples in the historical experience cache pool reaches the preset sample size threshold, a batch of historical samples are randomly sampled from the historical experience cache pool to perform gradient updates on the network parameters of the unloading decision model. This enables the unloading decision model to adaptively generate task unloading and resource allocation decisions in subsequent time slots based on changes in the user terminal local task queue, fine-grained remote task queue, edge server congestion level, and edge server energy consumption status.

10. A mobile edge computing task offloading and resource allocation system that implements the mobile edge computing task offloading and resource allocation method as described in any one of claims 1-9, characterized in that, include: The task acquisition module is used to acquire the task feature vector of the current time slot in a multi-edge mobile computing environment; The state awareness module communicates with the task acquisition module and is used to acquire the environmental state feature vector of the current time slot; The decision generation module, which is connected to the state awareness module, is used to input the state vector of the current time slot, which is constructed based on the task feature vector and the environmental state feature vector of the current time slot, into the preset unloading decision model to generate at least one candidate unloading decision for the current time slot. The execution path distribution module divides the tasks to be processed into local execution tasks and unloading execution tasks based on the candidate unloading decision. It sends the local execution tasks to the local computing resource allocation module and the unloading execution tasks to the uplink evaluation module and the edge CPU joint allocation module. The uplink evaluation module communicates with the execution path distribution module. It is used to construct the offloading utility coefficient based on the pairing relationship between the user terminal corresponding to the offloading execution task and the target edge server, and to obtain the uplink transmission evaluation result corresponding to each candidate offloading decision. The local computing resource allocation module communicates with the execution path distribution module and is used to obtain the amount of local processing tasks that the user terminal can complete in the current time slot and the corresponding local computing energy consumption based on the local execution tasks in the current candidate unloading decision. The edge CPU joint allocation module communicates with the execution path distribution module and is used to explicitly allocate the CPU computing resources of each edge server after the unloading task is uploaded to the target edge server for the current candidate unloading decision. It determines the CPU resource allocation result of the edge server in the current time slot and obtains the processing volume and CPU energy consumption of the unloading task completed by the edge server based on the CPU resource allocation result of the edge server in the current time slot. The comprehensive evaluation and execution module is connected to the uplink evaluation module, the local computing resource allocation module, and the edge CPU joint allocation module, respectively. It is used to calculate the reward value of each candidate offloading decision based on the comprehensive evaluation index set corresponding to each candidate offloading decision and through a preset comprehensive reward function. Based on the reward value, it determines the final offloading decision of the current time slot and executes the offloading task. The strategy update module communicates with the comprehensive evaluation and execution module to update the unloading decision model based on the actual unloading task execution results collected in the current time slot.

Images