CN-122027472-A - Combined optimization method and system for two-layer UAV-MEC network

CN122027472ACN 122027472 ACN122027472 ACN 122027472ACN-122027472-A

Abstract

The invention discloses a joint optimization method and a system for a two-layer UAV-MEC network, wherein the method comprises the steps of collecting historical requests and predicting content requests to obtain content popularity, enabling an upper unmanned aerial vehicle to select cache and plan according to the content popularity on a long time scale, placing the cache and plan into an experience pool, enabling a lower unmanned aerial vehicle to conduct local cache adjustment and unloading selection according to the content popularity on a short time scale, placing the cache and unloading selection into the experience pool, introducing CVaR, combining time delay and energy consumption to form rewards and constructing a unified target, intensively training and updating a plurality of agent strategies according to the experience pool and the unified target, and then issuing the agent strategies to each unmanned aerial vehicle for execution. The system comprises an upper unmanned aerial vehicle serving as an air small cloud server and a lower unmanned aerial vehicle serving as an air base station. The invention minimizes the average delay, the total energy consumption and the tail delay risk of the system through the collaborative decision of the two time scales. The method and the device can be widely applied to the field of mobile edge calculation.

Inventors

JIANG KANYANG
CHEN CI
YANG CHAO

Assignees

广东工业大学

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (8)

1. The joint optimization method for the two-layer UAV-MEC network is characterized by comprising the following steps of: Predicting the content request to obtain content popularity; The upper unmanned plane performs cache planning and track optimization according to the content popularity and puts the content popularity into an experience pool; The lower unmanned aerial vehicle carries out local cache adjustment and task unloading decision according to the content popularity and puts the decision into an experience pool; CVaR is introduced, and a unified target is constructed by combining time delay and energy consumption; and according to the experience pool and the unified target, intensively training and updating the multi-agent strategy, and then issuing to each unmanned aerial vehicle for execution.
2. The joint optimization method for a two-layer UAV-MEC network according to claim 1, wherein the step of predicting content requests to obtain content popularity specifically includes: Taking the content and the user as graph nodes, encoding adjacent relations into weighted edges, performing space aggregation and time evolution on the weighted edges, and outputting the next time slot request prediction; The output results are arranged in sequence to obtain the ranking; and using Zipf distribution as a mapping from the ranking to the probability to obtain the popularity of the content.
3. The joint optimization method for a two-layer UAV-MEC network according to claim 2, wherein the step of using Zipf distribution as a mapping from ranking to probability is expressed as follows: Wherein, the Representing time slots Content of the content Popularity probability of (2); representing content after the content is sequenced from high to low according to the request strength according to the request prediction result of the content in the next time slot The corresponding ranking; representing Zipf parameters; representing a total number of selectable content summarized by the content library; representing content number and ; Index variable representing content number used in denominator summation 。
4. The joint optimization method for a two-layer UAV-MEC network according to claim 2, wherein the calculation formula of the delay is expressed as follows: Wherein, the Representing a time slot; Representing time slots The total time delay of the internal system is calculated, Representing a user Is in time slot The transmission delay time generated in the process is equal to the transmission delay time generated in the process, Representing a user Is in time slot The internally generated computation delay; representing underlying unmanned aerial vehicle index and , Representing the total number of unmanned aerial vehicles at the lower layer; representing upper unmanned aerial vehicle index and , Representing the total number of unmanned aerial vehicles on the upper layer; representing user serial numbers covered by lower unmanned aerial vehicle and , Represented by the first The first layer of unmanned aerial vehicle coverage A name user identification; And (3) with Respectively summing index variables; Representing the user under the offload path With lower floor unmanned aerial vehicle The amount of task input data transmitted between them, Representing time slots Inner user With lower floor unmanned aerial vehicle A transmission rate therebetween; Representing the user under the offload path With upper unmanned aerial vehicle The amount of task input data transmitted between them, Representing time slots Inner user With upper unmanned aerial vehicle A transmission rate therebetween; indicating variables for binary uninstallation when the user Is in time slot Unloading to lower unmanned aerial vehicle Taking 1 if not, taking 0 if not; indicating variables for binary uninstallation when the user Is in time slot Unloading to upper unmanned aerial vehicle Taking 1 if not, taking 0 if not; representing the number of CPU cycles required for a task to execute on the lower unmanned side, Representing the number of CPU cycles required by the task to execute on the upper unmanned side; 、 and respectively representing available calculation forces of the lower unmanned aerial vehicle and the upper unmanned aerial vehicle.
5. The joint optimization method for a two-tier UAV-MEC network of claim 4, wherein the energy consumption is calculated as follows: Wherein, the Representing time slots Is used for the total energy consumption of the system, Represent the first The lower unmanned aerial vehicle is in time slot Is used for the energy consumption of the (a), Represent the first The upper unmanned aerial vehicle is in time slot Energy consumption of (2); Represent the first The lower unmanned aerial vehicle is in time slot Is used for the calculation of the energy consumption of the (c), Represent the first The upper unmanned aerial vehicle is in time slot Is calculated by the energy consumption; Energy consumption coefficient representing unit calculation time length; And (3) with Respectively represent the first The first layer of unmanned aerial vehicle coverage Name user The task of the system is unloaded to the lower unmanned aerial vehicle and the upper unmanned aerial vehicle; Represent the first The upper unmanned aerial vehicle is in time slot Is used for the cache energy consumption of the (a), Represent the first The lower unmanned aerial vehicle is in time slot Is not needed; representing upper layer unmanned aerial vehicle pair content If in time slot Content is processed The new transfer into the upper unmanned aerial vehicle cache takes 1, otherwise, taking 0; representing underlying drone pair content A binary permutation indication of (a); Represents the unit content energy consumption from the far Yun La to the upper unmanned aerial vehicle, And the unit content energy consumption constant of the content migrated to the lower unmanned plane is represented.
6. The joint optimization method for a two-tier UAV-MEC network of claim 5, wherein CVaR is defined as: Wherein, the Representing a random cost variable that is a function of the random cost variable, Represents the VaR threshold corresponding to the risk level, Represents a risk level and , Representing excess cost exceeding threshold and ; Representing a mathematical expectation operator.
7. The joint optimization method for a two-tier UAV-MEC network of claim 6, wherein the unified objective expression is as follows: Wherein, the The weight coefficient representing the energy consumption term, Represents the weight coefficient of the tail risk item and satisfies 、、 ; Representing the average level of the total time delay of the system and being used for representing the total time delay performance of the system; Represents the normalized reference value of the time delay, ; Representing an average level of total energy consumption of the system; Represents the normalized reference value of the energy consumption, ; Representing the conditional risk value of the total time delay of the system at the risk level.
8. A joint optimization system for a two-tier UAV-MEC network, configured to perform a joint optimization method for a two-tier UAV-MEC network as claimed in any one of claims 1-7, comprising: An upper unmanned aerial vehicle serving as an air small cloud server and a lower unmanned aerial vehicle serving as an air base station.

Description

Combined optimization method and system for two-layer UAV-MEC network Technical Field The invention relates to the field of mobile edge computing, in particular to a joint optimization method and a system for a two-layer UAV-MEC network. Background With the fusion evolution of the mobile internet and the internet of things, terminal-side services continue to develop towards high computing power, strong data and low time delay directions, and applications such as AR/VR, high-definition video enhancement, target detection, intelligent inspection, vehicle-road collaborative perception and the like are continuously emerging. On one hand, the application requires near real-time processing of the general data, and on the other hand, the application is subject to natural constraint of a battery, heat dissipation and form of terminal equipment, and delay and energy efficiency are difficult to be achieved by relying on local calculation alone. Mobile Edge Computing (MEC) has become an important path to improve the end-to-end experience by deploying computational power and storage resources on the side near the data source, reducing queuing and congestion from core network back and forth transmissions. However, in sparse ground infrastructure or disaster prone areas (e.g., mountainous areas, islands, deserts, and post-disaster sites), the fixed edge nodes are costly to build and maintain, long to deploy, and limited in coverage flexibility. An Unmanned Aerial Vehicle (UAV) carrying computing and caching capabilities can lift off as required to quickly form an air supplementary point coverage, wherein a two-layer UAV-MEC network is generally adopted, a lower UAV is close to a user to provide access and primary computing/caching, and an upper UAV serves as a 'mobile small cloud', and takes responsibility of high-capacity caching, computing and re-supplementing, relaying and the like. Around the above system, the prior art generally adopts "delay-energy consumption" weighted optimization as an overall goal, in an effort to minimize the average delay of the system and maintain acceptable energy overhead while meeting constraints such as capacity, unique load shedding, and task time limits. However, the average time delay is difficult to describe the extreme experience of 'worst several minutes', and factors such as link deep fading, user burst concentration or short distance from a hot spot area by an upper UAV (unmanned aerial vehicle) can cause a small amount of task time delay to rise sharply, so that SLA default or service perception cliff sliding is triggered. Therefore, the scheme cannot be adapted when facing to a real deployment scene. Disclosure of Invention In view of this, in order to solve the technical problem that the existing mobile edge joint method is mostly only aimed at minimizing the average time delay of the system, and thus cannot be suitable for practical application scenarios, in a first aspect, the invention provides a joint optimization method for a two-layer UAV-MEC network, which includes the following steps: collecting historical requests and predicting content requests to obtain content popularity; on a long time scale, the upper unmanned aerial vehicle selects a cache and a plan according to the popularity of the content, and puts the cache and the plan into an experience pool; On a short time scale, the lower unmanned aerial vehicle performs local cache adjustment and unloading selection according to content popularity and puts the content popularity into an experience pool; CVaR is introduced, and the time delay and the energy consumption are combined to form rewards and construct a unified target; and according to the experience pool and the unified target, intensively training and updating a plurality of agent strategies, and then issuing to each unmanned aerial vehicle for execution. In a second aspect, the present invention further proposes a joint optimization system for a two-tier UAV-MEC network, the system comprising: The system deploys upper unmanned aerial vehicle and lower unmanned aerial vehicle to serve as an air-in-air small cloud and an air base station respectively, and adopts two time scale operation frames with long period and short time slot combined. When a user initiates a calculation task, firstly, the lower unmanned aerial vehicle makes a decision to unload the calculation task to a local unmanned aerial vehicle, an adjacent unmanned aerial vehicle or an upper unmanned aerial vehicle for execution according to the local cache and the real-time link condition, so that high time delay caused by returning to a remote cloud is avoided. On a short time slot scale, the lower unmanned aerial vehicle quickly adjusts local caching and unloading strategies to adapt to user movement and request changes, and on a long period scale, the upper unmanned aerial vehicle synchronously updates the flight track and the global caching layout of the upper unmanned aerial vehicle so as to