CN-122019078-A - Cloud-edge collaborative asynchronous task scheduling and computing resource joint optimization method

CN122019078ACN 122019078 ACN122019078 ACN 122019078ACN-122019078-A

Abstract

The invention relates to a cloud edge collaborative asynchronous task scheduling and computing resource joint optimization method which comprises the steps of acquiring asynchronously arrived task information, edge equipment state information and server resource state information in a cloud edge collaborative computing network in real time, calculating priority weight of each task in an operation period according to time characteristics and computing requirements of the task information, performing preemptive reordering on a task queue based on the priority weight, constructing a system state vector based on the task information, the priority weight, the edge equipment state information and the server resource state information, generating a mixed action through a pre-trained deep reinforcement learning model, and executing scheduling of the task and distribution of computing resources in the operation period based on the mixed action. According to the method, the joint optimization of asynchronous task scheduling and resource allocation is realized through deep reinforcement learning, and the performance of the cloud edge collaborative computing system is remarkably improved.

Inventors

SU ZHIGUANG
ZHONG JING
LIU DERAN
ZHANG XIAOPENG
GUO SHUAI
SU XUAN
GUO BIN
ZHU SHANGQING
WU ZHENGZHONG

Assignees

中国人民解放军61618部队

Dates

Publication Date: 20260512
Application Date: 20251230

Claims (10)

1. A cloud-edge collaborative asynchronous task scheduling and computing resource joint optimization method is characterized by comprising the following steps of, In a cloud edge cooperative computing network, task information, edge equipment state information and server resource state information which arrive asynchronously are obtained in real time; According to the time characteristics and calculation requirements of the task information, calculating the priority weight of each task in the running period, and executing preemptive reordering on the task queue based on the priority weight; Based on task information, priority weight, edge equipment state information and server resource state information, a system state vector is constructed, a hybrid action is generated through a pre-trained deep reinforcement learning model, and scheduling of tasks and allocation of computing resources are executed in a running period based on the hybrid action.
2. The cloud-edge collaborative asynchronous task scheduling and computing resource joint optimization method of claim 1, wherein the acquiring asynchronously arriving task information, edge device state information and server resource state information in real time comprises, When the system is started, initializing a cloud edge cooperative computing network, and modeling asynchronous task arrival by using Poisson distribution, wherein the task arrival interval time delta Obeying an exponential distribution Exp (lambda), wherein k is a task, and lambda is an average arrival rate; Dividing the operating period T into n+2 time slots by a time division multiple access mechanism includes: A time slot for the edge device to acquire initial energy from the mobile edge computing server through wireless energy transmission; A time slot comprising a plurality of sub-time slots for task offloading; each edge device unloads tasks to an edge server through a wireless channel in the allocated sub-time slots, and each sub-time slot is only allocated to one edge device to generate an initial task queue; Time slots To the point of The method is used for processing tasks in parallel by the edge server and the cloud server and updating task state information and resource occupation conditions in real time; Time slots The method is used for summarizing task execution results of the edge server and the cloud server, and comprises task completion time And the number of completed tasks Wherein N is the number of preset parallel processing time slots; acquiring edge equipment state information at a fixed sampling frequency, wherein the edge equipment state information comprises an energy state and a channel state Wherein Representing the channel gain between edge device i and edge server e, The energy state comprises the current energy level and the energy acquisition rate of the edge equipment; Recording remaining computing resources of edge servers Remaining computing resources of cloud server 。
3. The cloud-edge collaborative asynchronous task scheduling and computing resource joint optimization method of claim 2, wherein computing the priority weight of each task comprises, Time slots In, based on the initial task queue, the arrival time of the task k Residual deadline And calculating the degree of urgency Performing normalization processing to generate arrival time factors Factor of remaining time And an emergency factor The formula is: , , , wherein, T is the operation period, For the task processing time threshold value, Is the maximum degree of urgency; The priority weights are calculated by weighted summation, the formula is: , Wherein, the As the priority weight of task k, 、、 Is a weight coefficient.
4. The cloud-edge collaborative asynchronous task scheduling and computing resource joint optimization method of claim 3, wherein performing preemptive reordering of task queues based on priority weights comprises, In time slot To the point of When the priority weight of a newly arrived task exceeds the priority weight of a currently executed task and the difference value is larger than a preset threshold delta P, interrupting the currently executed task, and storing the calculation progress of the currently executed task into a memory queue; And reordering the task queues added with the newly arrived tasks in a descending order according to the priority weights, generating an optimized task queue, and storing the corresponding priority weights.
5. The cloud-edge collaborative asynchronous task scheduling and computing resource joint optimization method of claim 4, wherein constructing a system state vector, generating a hybrid action via a pre-trained deep reinforcement learning model comprises, In time slot To the point of In, build system state vector The system state vector Including task information Channel state Remaining computing resources of edge server and cloud server And the remaining time of the operation period T Wherein, the method comprises the steps of, The amount of task computation for task k, Task resource requirements for task k; the system state vector is set Generating hybrid actions by inputting pre-trained deep reinforcement learning model Wherein t is the current slot index; the mixing action Comprising the steps of (a) a step of, User scheduling decisions For determining time slots An edge device i for offloading tasks in sub-slot n of (1), wherein, Indicating that device i is in a slot The nth sub-slot within it offloads task k, Representing idle, the deep reinforcement learning model is based on a system state vector Priority weights in (b) Preference for high from an optimized task queue Is assigned to device i; Offloading target decisions For determining the processing position of task k, wherein, Indicating that task k is processed at the edge server, Indicating that the channel is shunted to the cloud server, and when the channel gain is increased Above a predetermined threshold or remaining computing resources When sufficient, the deep reinforcement learning model is based on a system state vector Priority weights in (b) And edge server state, preferably high An edge server is allocated for task k; Computing resource allocation decisions Computing resources for allocating edge servers or cloud servers to task k and satisfying maximum computing resource constraints And ; The user scheduling decision and the unloading target decision are generated by adopting an epsilon-greedy strategy, optimal actions are selected according to the probability of 1 epsilon, random exploration is carried out according to the probability of epsilon, the computing resource allocation decision outputs a mean value through actor networks and overlaps Gaussian noise generation, and the allocation resources are ensured not to exceed the residual resources of the edge server or the cloud server.
6. The cloud-edge collaborative asynchronous task scheduling and computing resource joint optimization method of claim 5, wherein the training of the deep reinforcement learning model comprises, In the interaction with cloud edge collaborative computing network, in time slot To the point of Internal execution action After that, a state transition tetrad is generated And storing the data into an experience pool, wherein, Is a reward function; Randomly sampling a preset number of samples from the experience pool when the number of samples in the experience pool is not smaller than a preset threshold value; calculating a time sequence difference error based on the sampling sample and the Belman equation, and updating critic the network to minimize the mean square error loss; And updating actor the network by adopting a strategy gradient rising method, and optimizing the selection strategy of the mixed action until the model converges or reaches the preset training step number.
7. The cloud-edge collaborative asynchronous task scheduling and computing resource joint optimization method according to claim 6, wherein the deep reinforcement learning model is optimized based on a reward function, and the formula is: , Wherein, the In order to pay for the energy consumption, For time slots To the point of The total energy consumption of the task completed by the inner edge server or the cloud server; For the purpose of task timeliness rewards, For time slots To the point of The actual completion time of each task is within, Is a specified maximum delay; Rewards are given for the utilization of the resources, For time slots To the point of The total amount of computing resources actually used, Calculating a maximum amount of resources for all servers; a base prize value for task completion, C a base prize value for task completion, For time slots To the point of The number of tasks completed in; 、、、 Is a weight coefficient.
8. The cloud-edge collaborative asynchronous task scheduling and computing resource joint optimization system is characterized by comprising a data acquisition module, a reordering module and a task scheduling and resource distribution module; the data acquisition module is used for acquiring asynchronously-arrived task information, edge equipment state information and server resource state information in real time in a cloud-edge cooperative computing network; The reordering module is used for calculating the priority weight of each task in the running period according to the time characteristics and calculation requirements of the task information, and performing preemptive reordering on the task queue based on the priority weight; The task scheduling and resource allocation module is used for constructing a system state vector based on task information, priority weight, edge equipment state information and server resource state information, generating a mixed action through a pre-trained deep reinforcement learning model, and executing task scheduling and computing resource allocation in a running period based on the mixed action.
9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor implements the steps in the cloud-edge collaborative asynchronous task scheduling and computing resource joint optimization method according to any one of claims 1-7 when executing the program.
10. A storage medium storing a computer program, wherein the computer program when executed by a processor implements the steps of the cloud-edge collaborative asynchronous task scheduling and computing resource joint optimization method according to any one of claims 1-7.

Description

Cloud-edge collaborative asynchronous task scheduling and computing resource joint optimization method Technical Field The invention belongs to the technical field of cloud computing and edge computing, and particularly relates to a cloud-edge collaborative asynchronous task scheduling and computing resource joint optimization method. Background With the rapid development of 5G networks, internet of things and artificial intelligence technologies, demands of application scenes such as automatic driving, augmented reality and the like on task instantaneity and computation are increasing. The delay caused by long-distance data transmission of the traditional cloud computing architecture is difficult to meet millisecond-level response requirements, and the Mobile Edge Computing (MEC) is used for sinking computing capacity to a position close to the terminal equipment by deploying computing nodes at the network edge, so that network delay is greatly reduced, and backbone network pressure is relieved. However, the limited computing resources of MEC limit its ability to handle complex tasks, and cloud-edge collaborative computing becomes a key technology to solve the contradiction by combining low latency characteristics of edges and strong computing power of cloud. Currently, cloud edge collaboration faces multiple challenges that task arrival time is random and asynchronous, edge equipment relies on an energy collection technology to supply power, energy states dynamically change, computing resources of an edge server are limited, and cloud processing is limited by communication delay. In addition, the service quality requirements of tasks are differentiated, for example, real-time tasks need edge-first processing, and computationally intensive tasks depend on cloud computing power. The traditional synchronous scheduling method needs to wait for all tasks to reach for post-processing, so that resource waste and overtime of high-priority tasks are caused, and the traditional optimization algorithm has high computational complexity and is difficult to adapt to task asynchronism and equipment energy dynamic change. Therefore, an intelligent and flexible scheduling and resource allocation method is needed to realize efficient optimization of cloud-edge coordination. Disclosure of Invention The invention aims to provide a cloud-edge collaborative asynchronous task scheduling and computing resource joint optimization method, which aims to solve the problems of low task scheduling and resource allocation efficiency caused by asynchronous task arrival, dynamic change of edge equipment energy and resource limitation in the existing cloud-edge collaborative computing network. In order to achieve one of the above objects, an embodiment of the present invention provides a cloud-edge coordinated asynchronous task scheduling and computing resource joint optimization method, the method comprising, In a cloud edge cooperative computing network, task information, edge equipment state information and server resource state information which arrive asynchronously are obtained in real time; According to the time characteristics and calculation requirements of the task information, calculating the priority weight of each task in the running period, and executing preemptive reordering on the task queue based on the priority weight; Based on task information, priority weight, edge equipment state information and server resource state information, a system state vector is constructed, a hybrid action is generated through a pre-trained deep reinforcement learning model, and scheduling of tasks and allocation of computing resources are executed in a running period based on the hybrid action. As a further improvement of an embodiment of the present invention, the real-time acquisition of the asynchronously arrived task information, the edge device status information and the server resource status information includes initializing a cloud-edge cooperative computing network and modeling the asynchronously arrived task by poisson distribution when the system is started, wherein the task arrival interval time deltaObeying an exponential distribution Exp (lambda), wherein k is a task, and lambda is an average arrival rate; Dividing the operating period T into n+2 time slots by a time division multiple access mechanism includes: A time slot for the edge device to acquire initial energy from the mobile edge computing server through wireless energy transmission; A time slot comprising a plurality of sub-time slots for task offloading; each edge device unloads tasks to an edge server through a wireless channel in the allocated sub-time slots, and each sub-time slot is only allocated to one edge device to generate an initial task queue; Time slots To the point ofThe method is used for processing tasks in parallel by the edge server and the cloud server and updating task state information and resource occupation conditions in real time; Time slots T